Java Performance On Ubuntu Vs. Windows Vista

Henckle writes "Phoronix did a comparison of the Java performance between Ubuntu and Windows Vista. They tested both Java and OpenJDK on Ubuntu 8.10 and Java on Windows Vista Premium SP1, all with stock configurations. To no-one's surprise, Ubuntu was faster in a majority of the tests. The two OSs were similar in ray-tracing, and Vista was faster at Java OpenGL due to shortcomings with the Linux graphics driver."

Comments

I must say...

[Thelasko]

that new 64-bit Java plugin for Linux is smokin! No waiting for applets to load or anything.

Re:I must say...

[techprophet]

Yeah I'm using it too. Fast now has a new definition.

Interesting, but lacking some crucial details...

[Smidge207]

Just out of professional curiosity: what was the partition layout on the laptop? As benchmarks in some articles have shown, the early part of a drive is faster than later part (sequential transfer rate). With a constant areal density, data flies under the read/write heads faster on the outer larger-radius tracks.

This is a something that's hard to get right when benching win vs. lin on the same HW, since usually you have a fairly normal dual-boot install, and one has the advantage of the outer tracks.

It's probably not a big deal if you have two adjacent 10 or 15GB partitions, with a big data partition somewhere else.

Ideally you'd re-partition and run benchmarks with each system installed to the first few GB. To get reasonable numbers for I/O dependent tests, you could make a scratch partition that you reformat to ext3 or ntfs before running the tests. Then have I/O benchmarks do their work in that scratch partition). (or XFS, see my previous posts for XFS tuning . XFS's delayed allocation means it doesn't start writing until you runs low on RAM, or it otherwise decides it's time to start. This means uninterrupted reading for longer = less seeks = faster.) This tests fresh filesystems, not somewhat worn filesystems that everyone will actually have after even a day of use, but usually it's not a big difference because most filesystems don't suck that badly when they're not close to full.

I thought Vista SP1 was supposed to fix slow file I/O. Oh, IIRC, that was just slow file copying when you do it via the GUI shell. So never mind, I guess either your partitioning really skewed things in favour of GNU/Linux, Vista sucks at the file-encryption workload, or it was CPU-limited and the older JVM on Vista loses on that code.

Oh, well, sorry for the rambling, just my $1.00-.98....

=Smidge=

that's odd

[z-j-y]

those tests (CPU burners) should perform the same on Linux or Windows, I don't see why JVM would behave differently.

Re:that's odd

[Shados]

They -are- different JVM builds, so its possible (as is common in the JVM's history) that some bug fixes improve performance wildly... Not across the board though, so something's wrong, either with the JVM, or with Windows itself... but something is seriously messed up.

Re:that's odd

[Actually, I do RTFA]

those tests (CPU burners) should perform the same on Linux or Windows, I don't see why JVM would behave differently.

They left on Aero, and acutally spent more money to get the version of Vista with it. The graphics chip was the weakest integrated chip that, according to the manufacturer, supported Aero. In reality, no doubt, the Intel chip pushes work onto the CPU because we know what happens when a manufacturer claims that the hardware is just able to support a feature. Hell, even their website phrases the compatibility in weasly terms.

Re:that's odd

[enharmonix]

I wouldn't put it past MS

Come to think of it, I don't think I can blame this one on MS. If it's running in a VM, I think that's a setup issue...

Re:that's odd

[Dr.Merkwurdigeliebe]

If MS put it in a VM, then it's an MS issue ...

Re:that's odd

[Hal_Porter]

IE7 isn't 'run inside a VM' it's run in a process which is prevented from doing a lot of stuff, e.g. it is only allowed file access to the temp directory.

http://blogs.msdn.com/ie/archive/2006/02/09/528963.aspx
Internet-facing applications such as browsers are inherently at a higher security risk than other applications because they can download untrustworthy content from unknown sources. IE7's Protected Mode leverage's Windows Vista's UAC, MIC and UIPI features to boost browser security. In IE7's Protected Mode-which is the default in other than the Trusted security zone-the IE process runs with Low rights, even if the logged-in user is an administrator. Since add-ins to IE such as ActiveX controls and toolbars run within the IE process, those add-ins run Low as well. The idea behind Protected Mode IE is that even if an attacker somehow defeated every defense mechanism and gained control of the IE process and got it to run some arbitrary code, that code would be severely limited in what it could do. Almost all of the file system and registry would be off-limits to it for writing, reducing the ability of an exploit to modify the system or harm user files. The code wouldn't have enough privileges to install software, put files in the user's Startup folder, hijack browser settings, or other nastiness.

It's a good idea and it doesn't affect performance.

Re:that's odd

[thetoadwarrior]

The JVM still has to work with the OS to make things work so things are different between the the JVMs and performance can vary depending on how the underlying OS deals with certain things.

Re:that's odd

[Fweeky]

Java has quite a few tuning options; limits on heap and garbage pool sizes, different garbage collectors and strategies for running them, different optimization levels. Maybe it's defaulting to client mode on Windows, thus putting more effort into reducing memory use by running the gc before growing the heap (and placing stricter limits on its size and the sizes of the gc generation pools), doing less aggressive optimizations and putting them off later.

Java also has bit of a history of enjoying cheap syscalls; 1.5 ran somewhat poorly on FreeBSD because it called gettimeofday() at some significant rate of knots, and system calls on Windows are significantly more expensive than that, while Linux prides itself on them being very cheap.

Re:that's odd

[lacoronus]

The Java launcher has two VMs to choose between - the server and the client VM. Unless forced by command-line switches, it chooses automatically.

The client VM does a lot on an "as needed" basis - classloading, JIT compilation, etc. The server does most work upfront, before the app starts. This means client = faster startup and less resources consumed, server = faster execution and more resources consumed.

On Ubuntu Phoronix ran the OpenJDK Server VM, and got the same times as the stock VM. On Windows, they ran the stock VM and got worse performance. I know that on Windows, with a setup like the one they use, the launcher will always select the client VM.

I would be interested to know which VM (client or server) was used on the benchmarks. I strongly suspect they ended up with the server VM on Ubuntu and client on Windows.

Re:that's odd

[Cajun Hell]

First of all, I think it's interesting that someone would bother to use Java (instead of, say, C) for ray-tracing. I guess JVMs have come a long way.

those tests (CPU burners) should perform the same on Linux or Windows, I don't see why JVM would behave differently.

Yep, although if they had used a more interesting test machine (dual Opterons instead of a single CPU), then the different kernels (not JVMs) might have made a difference, thanks to NUMA-related issues.

Re:that's odd

[Anonymous Coward]

I remember Java benchmarks where I would run the program and get speeds 1.3x slower than C and other people would run it and get 4x slower than C. Same OS, same jdk, same benchmark.

Difference? They had an Intel CPU and I had an AMD. Hotspot did not at the time have any way to word-align variables on the stack and this caused a huge hit on Intel but not AMD (due to cpu designs at the time, probably not now).

Re:that's odd

[shutdown -p now]

MS doesn't make or even distribute the Java VM for Windows. It's all Sun's code (unlike the situation on OS X, where Apple controls the releases).

Re:that's odd

[immcintosh]

Go back a few posts. This is a thread about the possibility that Windows is running the Java VM inside a second sandbox VM, therefore an MS issue.

Re:that's odd

[enharmonix]

Protected Mode leverage's Windows Vista's UAC

You're right that its not running in a VM like VMware, but we're talking about the same thing. IE7's protected mode is just a UAC Virtualization applied to IE7. For more info on virtualized processes: http://www.windowsecurity.com/articles/Protecting-System-Files-UAC-Virtualization-Part1.html but it sounds like you are already familiar with it.

BTW, the link is for anybody else reading, it's apparent you are familiar w/ how UAC works. Cheers.

Re:that's odd

[shutdown -p now]

IIRC, Vista runs untrusted processes in their own VMs (IE7, for instance). I wonder if they did this to the JVM. Seems kinda silly to put a VM within another VM, but I wouldn't put it past MS.

Vista doesn't run IE7 in a VM. IE7 runs in a "sandbox", which is really just an extremely underprivileged special user account which only has the rights that are needed to download stuff from the net, and store and load it from the cache. It does not affect performance in any way. Also, it's an IE7 feature, not a Vista feature (Vista provides the more efficient means to do such sandboxing, which is why the feature is only enabled on Vista, but it's IE7 that uses that technique so far).

Re:that's odd

[Filip22012005]

I remember Java benchmarks where I would run the program and get speeds 1.3x slower than C and other people would run it and get 4x slower than C.

It's all dependent on rerouting power to the impulse engines.

Re:that's odd

[Hal_Porter]

Well it's virtualized in the sense that when a process writes to path, the kernel checks for a list of paths that the process can't write to because of Vista UAC but could have written to on XP, chucks a prefix on the front and passes the prefixed path down to the filesystem.

Describing that as a VM is a bit misleading though, because it sounds like you have a much larger performance hit than is actually the case. Vista has some bloaty design changes, but this isn't one of them. Actually it is quite sensible.

Re:that's odd

[nxtw]

The graphics chip was the weakest integrated chip that, according to the manufacturer, supported Aero. In reality, no doubt, the Intel chip pushes work onto the CPU because we know what happens when a manufacturer claims that the hardware is just able to support a feature. Hell, even their website phrases the compatibility in weasly terms.

The article mentions that the laptop uses the PM965, which is the GPU-less mobile chipset - they probably mean GM965. The GM965 isn't the weakest Intel GPU than can run Aero, either. Aero runs on the Intel 945-series GPUs and above. The 915 series has specific hardware limitations that prevent it from running Aero.

Aero on integrated graphics does use a lot of memory bandwidth. I've seen degraded Aero framerates on Intel GPUs using single channel memory.

The chip itself can't "push any work onto the CPU". The driver could, but there is no evidence of the Intel graphics driver using significantly higher CPU than than competitors' drivers while running Aero. At any rate, there shouldn't be a significant hit in CPU time or memory transfers when nothing is changing on the screen..

Re:that's odd

[ChrisA90278]

The two JVMs are different, and actually use different logic inside. One has to use the POSIX APIs while the other is using Windows APIs to get to OS services like reading files and drawing to the screen.

But you are correct if the test is a simple compute loop. One would think it should be the same. But no not really because one JVM was built with gcc and the other with some Microsoft compiler. Simply upgrading gcc might change the result. The JVMs really are different

Re:that's odd

[Anonymous Coward]

They *also* left on Ubuntu's desktop effects. So I would say this is a fair comparison. Beryl gives the same effects while not slowing the computer to a crawl, this is absolutely a fair test.
          It'd be interesting, however, to turn off Beryl, turn off Aero, and rerun the tests.

Re:that's odd

[Bert64]

Kinda like chroot, which unix has had for a long time...

Re:that's odd

[Your.Master]

Not really, it behooves Sun to make their code so that it doesn't fall into an untrusted process VM if they don't want it.

That said, the point is moot because what really happens is IE7 and up on Vista runs as a Low process which has VERY limitied filesystem and registry access, with essentially no performance hit; and anyway Java doesn't run on that. Chrome does something very similar, for what it's worth, but the other browsers choose to run as user processes.

Re:that's odd

[bckrispi]

Exactly. When it's time to get down to the literal wire, the JVM has to make native system calls to get its job done.

Re:that's odd

[shutdown -p now]

Kinda like chroot, which unix has had for a long time...

It's actually a bit more than that. It's not just limited restrictions to access files & registry - they actually lock down the process completely; for example, it is not able to send WMs to other processes, or in general communicate with them using any other form of IPC, except for the one single parent process, and only through channels explicitly designated by that parent process for communication.

Here is a more detailed technical description of how all this works, and what it gains.

Re:that's odd

[Bert64]

chroot will achieve a similar effect, for things like unix sockets, fifos and potentially device nodes which are not reachable from inside of the chroot...
On unix pretty much everything is a file, so locking down access to files is all you really need to do. It's the simplicity which makes it powerful...
On windows you cant just lock down access to files, you have to lock down tons of other things too, and complexity very much gets in the way of security.

Re:that's odd

[lukas84]

It's not as easy as you're trying to make it.

IE7 is an interactive graphical application.

Try running e.G. Konqueror in chroot, and restrict it from interacting with other programs in the same X session.

That's my laptop!

[enharmonix]

For this round-up we had used a Dell Inspiron 1525 notebook (PM965 + ICH8M Chipset) with an Intel Core 2 Duo T5800 processor clocked at 2.0GHz, 3GB of DDR2 memory, 250GB Hitachi HTS543225L9A300 HDD, integrated Intel 965 graphics, and a screen resolution of 1280 x 800.

I'm shopping for a new laptop and I settled on the 1525 with this exact configuration. Ironically, I can't decide if I want Ubuntu or Vista Home Premium on it. I don't really use much Java so this article doesn't really help me... What a weird coincidence, though.

Re:That's my laptop!

[neuromanc3r]

Ironically, I can't decide if I want Ubuntu or Vista Home Premium on it.

It's not like the two are mutually exclusive...

Re:That's my laptop!

[Dr.Merkwurdigeliebe]

I do not see coincidence, I see providence. I see purpose.

Re:That's my laptop!

[Darkness404]

Depends what you want. Vista will at least be somewhat usable on that configuration so unless you either don't know how to install Ubuntu, have XP that you want to run in a VM rather than dual-booting Vista, or want to save $50 you could always go with Vista because installing Ubuntu over Vista is trivial, and the hardware is going to be rather easy to work with Linux (no nVidia drivers, etc.) and the only other thing that Dell really installs for you is libdecss and other codecs which aren't too hard to find elsewhere. However, installing Vista over Ubuntu in a dual-booting mode is going to be a total pain requiring the reinstall of GRUB and perhaps changes to the partition layout.

Re:That's my laptop!

[Anonymous Coward]

You're having a hard time picking between Linux or Windows? Go Linux, it's cheaper to change your mind.

Re:That's my laptop!

[Dishevel]

You're having a hard time picking between Linux or Windows? Go Linux, it's cheaper to change your mind.

This is actually a really good point. Mod Parent UP.

Re:That's my laptop!

[rundgren]

NTFS support on Ubuntu is really pretty good. The performance is not quite as good as native, but I think saying that "neither works particularly well in their non-native environments" is going to far, as functionality and stability is very good. You can even run Ubuntu easily on NTFS with wubi. Check out this review: http://www.linux.com/feature/130713

Re:That's my laptop!

[Count Fenring]

Not, strictly speaking, true...

The NTFS-3g driver for Fuse works very well, at this point, and there are ext2/3 drivers for Windows if you look for them (Albeit the free one that I know about doesn't handle journaling in Ext3).

Here you go!

My personal favorite setup is having an ext3 home directory in Linux, and using the My Documents folder in NTFS as a media directory.

Slight annoyance of having two desktops, but other than that, pretty ideal.

Re:That's my laptop!

[Anonymous Coward]

I do not see coincidence, I see providence. I see purpose.

I see someone trying to sell more Dell Inspiron notebooks.

Re:That's my laptop!

[Anonymous Coward]

That's so crazy it just might work! Thanks. (Sorry to post anonymously but my karma is spiraling the drain right now :)

Re:That's my laptop!

[mR.bRiGhTsId3]

Not quite as good as native would seem to me to be tremendously overstate. I'm not an expert on FUSE architecture, but based on the fact its a Filesyste in USErspace, I'm going to go with, performance should suck horribly. Every filesystem operation requires 2x the number of context switches because the kernel has to switch back into userspace in order to run the FUSE vfs layer.

Re:That's my laptop!

[shaitand]

It doesn't. NTFS-3g is much faster than previous solutions but still isn't nearly as fast as using a native filesystem.

If your definition of 'media' includes 700mb-1gb+ files then forget it. If its word documents and powerpoint then by all means.

Re:That's my laptop!

[Anonymous Coward]

Burn that crap and install this. This is the real thing. I'm using this driver and it blows away ext2-IFS. It's bloody fast, almost as fast as a real Linux disk-access, and it supports ext3's journaling too.

ext2fsd - Download here.

As a bonus, instead of getting a festival of BSODs when you access an Unicode volume, ext2fsd fails gracefully. And, as an user-level program, if it crashes, it crashes. ext2-IFS is a kernel driver, and if it crashes... BSOD!

Re:That's my laptop!

[Count Fenring]

I'll check that out, although these days my windows use is down to games and nothing but, at least on my personal machine.

Re:That's my laptop!

[raduf]

Uh... actually it's not. Assuming you'll change your mind, if you start with with windows you have it installed from the start, and thus pay a small amount included in the price of the laptop. But if you start with linux, you'll have to buy a separate windows license which will most likely be more expensive.

So it's "cheaper _unless_ you change your mind".

Re:fast java

[calmofthestorm]

by cheating:-)

To be fair, you can usually get a bit of extra performance out of the windows version of superpi by running it in linux under wine than in windows.

Unrelated, but amusing.

always trust phronix to mess a benchmark up

[MancunianMaskMan]

they used java 1.6.0_10 on linux and 1.6.0_07 on windows. Hate to give the benefit of the doubt to ballmer & co but in spite of the minor version number, a lot of work in performance has been done on Java recently. The result is pretty meaningless.

Re:always trust phronix to mess a benchmark up

[hansamurai]

Excellent point, u10 was the next "major" minor release after u07, check out the release notes for it here, they're... long.

http://java.sun.com/javase/6/webnotes/6u10.html

Re:always trust phronix to mess a benchmark up

[Anonymous Coward]

Even better, take a look at the release notes for 07 vs 10... its a different version of the VM. (10 vs 11)

Re:always trust phronix to mess a benchmark up

[samkass]

Indeed, the u10 version of JDK6 is probably the biggest "point" release Java's ever had. For some strange reason Sun has decided not to use the last decimal point of their version numbers, but really this was JDK 1.6.1.

Any comparison of pre-u10 benchmarks with u10-or-later are pretty much completely invalid. This is especially true on Windows, which now uses hardware accelerated pipelines all over the place, so will probably be even more dramatically faster than Linux on the graphics tests.

Re:always trust phronix to mess a benchmark up

[jfim]

they used java 1.6.0_10 on linux and 1.6.0_07 on windows. Hate to give the benefit of the doubt to ballmer & co but in spite of the minor version number, a lot of work in performance has been done on Java recently. The result is pretty meaningless.

Indeed, this is a fairly lame benchmark. Java 6 update 10 is when they switched to the "consumer JRE" (also known as update "N"), which included a lot of changes to lower the footprint of the JRE, improve startup time, etc.

Re:always trust phronix to mess a benchmark up

[mhall119]

Yeah, that's a pretty bad comparison. Were the testers at Phronix completely unaware of the performance work Sun has been putting into update 10? They've only been talking about it for like a year.

Re:always trust phronix to mess a benchmark up

[Vellmont]

they used java 1.6.0_10 on linux and 1.6.0_07 on windows.

Yah, and they also tested with the new openJDK 1.6.0_0, which had nearly identical results as Suns JVM. The differences between Windows and Linux are NOT subtle, and you normally don't get 3x performance increases with mature code. So I have a hard time believing that these major performance differences are really from going from 07 to 10, especially when you factor in the results from openJDK.

I still think it'd be interesting to repeat the tests with the same JVM version on Vista and Ubuntu, but I'd hardly call the results "meaningless".

Re:always trust phronix to mess a benchmark up

[Yfrwlf]

"Whenever I hear the word activist, I reach for my revolver."

Tough crowd. You from China? :D

Ray tracing in Java

[olddotter]

Have computers or JIT compilers gotten fast enough that people actually do ray tracing in Java?

Re:Ray tracing in Java

[Yetihehe]

Yes. Sunflow is one example. I did my own tests, java vs c++ (almoste the same code) and java was only about 1.3 times slower.

Re:Ray tracing in Java

[shutdown -p now]

Compared to plain C/C++ code (no non-standard compiler intrinsics etc), Java has been fast enough for a long time now. Think about it: it's all just math, and JIT (which may itself be slow - it doesn't matter!) generates pretty much the same native code as a C compiler would. g++ can still do some trickier optimizations, which may account for the odd 5-10% of difference; but hardly more than that. .NET can actually fare even better, because it supports raw unsafe code and data pointers with arithmetic, dynamic allocation of stack memory (alloca), and unions - this essentially covers all optimization tricks available in ANSI C.

Re:Ray tracing in Java

[Anonymous Coward]

java was only about 1.3 times slower.

Pretty ambiguous language.

I assume you mean java was only about 1.3 times the speed of C++?

What you said was:
java = 2.3 x C++

but you could have meant:
java = 1.3 x C++

Re:Ray tracing in Java

[Waffle Iron]

Have computers or JIT compilers gotten fast enough that people actually do ray tracing in Java?

I can believe it could be. Java can be just as fast as C as long as you write code the way you'd write it in C. That means statically allocating your data in arrays and avoiding creating new objects (which results in allocation, constructor and GC overhead, as well as cache thrashing).

The problem most people have with Java performance is that most problems become very unwieldy if written in a C-like fashion, and everything in the Java language and culture encourages new object creation rather than reuse of mutable static objects. By the same token, the Java-like way is much easier to code and less prone to bugs than the C-like way. In a lot of applications, especially the ones where Java is heavily used, that's more important than performance.

The special case of raytracing is heavy on arrays and number crunching, and doesn't require much "business" type stuff like string processing or object abstractions, so coding in the C-like fashion is certainly feasible.

Re:Ray tracing in Java

[Anonymous Coward]

> I assume you mean java was only about 1.3 times the speed of C++?

This is Java we're talking about, so no.

1.3 times slower = .76 = 76% as fast as C++.

Re:Ray tracing in Java

[Anonymous Coward]

java doesn't have unsigned integers. Using unsigned integers can improve math performance significantly.

Re:Ray tracing in Java

[phantomfive]

Java is not much slower once it gets going in a loop (like ray-tracing, where it runs the same chunk of code over and over again), because once it is compiled it will run as assembly code.

In my experience, the thing that java does the worst on performance-wise is arrays: it has to check to make sure you don't overflow the array. This can slow down an array-element access as much as 10 times. I actually knew a guy who was so annoyed that he wrote a java pre-processor that converted his array into a sequence of variables (so if he coded ar[6] = 5; it converted it to ar_6 = 5). Pretty intense, but it was for a graphics package on a mobile phone, where a bit of speed CAN make a difference.

Re:Ray tracing in Java

[Yetihehe]

x - execution time of java program y - execution time of c++ program x = 1.3 * y I don't specifically remember exact results now and can't repeat it (java version already very modified). First execution of main program loop was nearly 3 times slower in java. Then after second execution it stabilised for next 98 executions (JIT worked after it had enough data on how to optimize code).

Re:Ray tracing in Java

[shutdown -p now]

Rightly spotted. With two's complement integers, the integer itself isn't really "signed" or "unsigned" - addition, subtraction and multiplication work exactly the same regardless of it, and you only need two distinct operations for signed and unsigned division. I don't know any good (i.e. performant) solution for Java unsigned division, though...

That's another place where .NET scores higher, by the way.

Re:Ray tracing in Java

[Anonymous Coward]

How can something be "1.3 times slower?" 1.3 times what?

Re:Ray tracing in Java

[Anonymous Coward]

How can something be "1.3 times slower?" 1.3 times what?

From the post-chain, I quote: "java vs c++ (almoste the same code) and java was only about 1.3 times slower"

Thus:
- Java is (in this instance) compared to C++
- Java is "only about 1.3 times slower" than C++
- Although not ideally worded, the above translates to:

SpeedOf(Java) = SpeedOf(C++) / 1.3

or

SpeedOf(C++) = 1.3 * SpeedOf(Java)

One result of many, confirming what most who do serious programming, development and deployment work, already know:

The speed of Java is more than sufficient. Particularly so in server contexts.

It might not be the fastest. It doesn't matter. It's fast enough.

I hope this helped.

Re:Ray tracing in Java

[pjt33]

How often do you need to do unsigned integer division? I've been programming for 20 years without once wanting it.

Re:Ray tracing in Java

[mR.bRiGhTsId3]

In my understanding java can go further and actually optimize layout in code hot spots to increase locality of reference, potentially giving you even more performance on certain workloads.

Re:Ray tracing in Java

[Anonymous Coward]

What compiler where you using for the C++ code? If it's GCC, your results are meaningless. Try a real compiler, with optimizations enabled and targeting at least the i586, and you should see that the C++ code performs significantly faster than the Java code.

The JIT in Java has made advancements in optimizing code in recent years - but so have C/C++ compilers. The code generated by GCC is notoriously poor. Try the Intel compiler to get more meaningful results.

Re:Ray tracing in Java

[Bert64]

Usually means that something else was 1.3 times the speed of the slower one, therefore the slower one was 1.3 times slower than the faster one... It's an incorrect way to say it, but that's usually what people mean.

Re:Ray tracing in Java

[chthonicdaemon]

I think 'meaningless' is quite a stretch. The Intel compilers are definitely better on Intel chips, but GCC is accessible and generates good code for other chips too. It is more of a known quantity for the kind of person who reads Slashdot. The reality is that C++ is _not_ the gold standard for speed any more than GCC (or Intel's compiler) is a standard for compilers. I think that what we are seeing is the convergence of compiler tech. No language is inherently slower than any other language -- compilers are more or less efficient, languages may make their job easier or harder. In the end, language is not all that important, compilers are.

Re:Ray tracing in Java

[InfiniteLoopCounter]

How often do you need to do unsigned integer division? I've been programming for 20 years without once wanting it.

I have a somewhat obscure usage for you that you might not have thought of.

You can perform optimizations when using products of prime numbers (you can fit more in an unsigned number). For instance, to test if a number is missing from a given set, just see if the remainder is zero. Where this may be useful instead of just using bit operations (and where division comes in) is when you may have the same prime multiplied more than once. You can then check the remainder, if zero, do an unsigned divide of the original product and then check the remainder again (rinse & repeat).

Again, not likely to come up often. But this maybe useful in scenarios where many rows of numbers with some unknowns have to be optimized for space and/or you would like to omit some slower to process 'if' statements.

Re:Ray tracing in Java

[ultranova]

In my experience, the thing that java does the worst on performance-wise is arrays: it has to check to make sure you don't overflow the array. This can slow down an array-element access as much as 10 times.

What ? How the hell can a simple (i > -1 && i < array.length) take up 9 times as much time as the actual array lookup ? I have a very hard time believing this.

Re:Ray tracing in Java

[Trogre]

So only about 5.2 times slower than the same code written in C then :p

Fairness

[abigsmurf]

I love the way in every test test Vista loses it's "Ubuntu is faster" but in the test where Vista wins, they explain and excuse it going "bad opengl drivers".

Either give possible reasons for slow performance of one OS each time or don't do it at all. To excuse bad performances in a benchmark in such a selective manner reeks of bias. Who's to say the Vista performance gap wasn't caused by bad drivers? Indeed the a lot of the tests where it was slower are ones involving disk access and Vista is known for being slow at this sort of thing.

Re:Fairness

[Dr.Merkwurdigeliebe]

When "Vista is known for being slow at this sort of thing", that's a cause of the OS itself, which is different than something that the OS relies on. You can't say that Vista lost at some benchmark because it's a shitty OS, therefore, it's not fair to call it a shitty OS.

Re:Fairness

[Chirs]

Likely they mention the issue being the graphics drivers in linux because graphics drivers are a continuing sore spot.

As for disk access, in my experience the vast majority of disk access speed issues are not driver related, but are due to the I/O subsystem, elevator algorithms, and filesystem code in the OS.

Re:Fairness

[Vellmont]

I love the way in every test test Vista loses it's "Ubuntu is faster" but in the test where Vista wins, they explain and excuse it going "bad opengl drivers".

Maybe that's because Ubuntu is an open source application where we actually know why the test gave bad performance, and actually know that it's going to improve in the future?

Who's to say the Vista performance gap wasn't caused by bad drivers?

Don't dismiss the advantages of an open system where you can actually understand what's under the hood as just "test bias".

Re:Fairness

[bigpaperbag]

I think his point was that the drivers are a known issue just like disk access is a known issue and both should be noted equally. Instead of one being noted, and the other not which implies a mysterious and general terribleness.

Re:Fairness

[PinkyDead]

...and when Microsoft have third party support ranging from acquiescence to mindless obedience to the exclusion of all others, complaining that it's a biased disadvantage in a benchmark is a bit rich.

Re:Fairness

[binarylarry]

Graphics drivers have been continually improving on Linux.

Intel, Nvidia and ATI hardware all works MUCH, MUCH better than it used to.

Not that things are perfect, but it's important to point out that things are getting better all the time.

Re:Fairness

[abigsmurf]

Sorry, that's just a stupid argument. Because an application is closed source they're never going to boost performance? Because you can't see the source code, you don't know if it's drivers causing bad performance?

You can see through benchmarking and general usage where an application is falling down in performance terms, is doesn't matter if you've access to course code or not. You say Vista will never have it's performance boosted? Have you seen the comparisons between a clean install and the latest updated version?

One of the most obvious examples of clear bad drivers in OS' are graphics drivers, regardless of if it's Vista or Linux, closed drivers or open drivers, it's usually plainly obvious if drivers are resulting in bad performances.

Besides which, benchmarks are for the here and now. OpenGL drivers may improve for Linux, Vista may improve its file handling, JavaVMs may see better optimisation, heck, even the damn benchmarks will probably get updated. None of this has anything to do with open or closed source.

Re:Fairness

[Ed Avis]

Maybe that's because Ubuntu is an open source application where we actually know why the test gave bad performance, and actually know that it's going to improve in the future?

I don't think you can assume that. Remember the recent benchmarks showing that Ubuntu had become slower in recent releases? Even if those were flawed, you still can't assume that future releases will be faster, any more than you could assume that future Windows releases will be an improvement. It's quite possible for software to get slower not faster.

Re:Fairness

[Vellmont]

Sorry, that's just a stupid argument. Because an application is closed source they're never going to boost performance?

No, because the OS is closed source you'll always be at the mercy of the license provider to boost performance. Honestly, how much does Microsoft _really_ care about Java performance? Not a hell of a lot.

Besides which, benchmarks are for the here and now. OpenGL drivers may improve for Linux, Vista may improve its file handling, JavaVMs may see better optimisation, heck, even the damn benchmarks will probably get updated. None of this has anything to do with open or closed source.

I guess I have to disagree here. None of those improvements happen in a vacuum. Microsoft decides whether to improve performance based on market decisions. The linux community makes decisions largely based on user needs. If someone really needed super-duper Java performance, they could go out and craft something that's soley performance based.

The point here is that closed source performance improvements are granted to you by the software company (The Great And Powerfull Oz!). Open source performance improvements can come from the guy who actually wants the improvements.

Re:Fairness

[chrb]

we actually know why the test gave bad performance

No we don't. If they had compared many different graphics cards on the same Linux system, and shown that there was some performance degradation on a particular card that occurred every time the test was run, then we might know why the test performed badly. At the moment the idea that it was the graphics drivers is just a hypothesis.

Re:Fairness

[Vellmont]

It's quite possible for software to get slower not faster.

Very true. I guess I was referring to this specific instance, where the article mentions the openGL layer of Mesa wasn't geared towards performance yet, but eventually will be.

Re:Fairness

[jellomizer]

By your argument Linux in theory should be light-years faster then Windows Mac OS or any other OS. However there are things that are slow in Linux. Why is Open GL so slow. Couln't you make MesaGL faster and more compatible with OpenGL.

You may be at the mercy of Microsoft yes Microsoft may not care about Java vs. its own .NET platform. However if Microsoft could whip Linux in Java Performance they would love it. Fine Vista loss these benchmark but the excusing the one area it did win was in very poor taste.

Re:Fairness

[lofoforabr]

It's quite possible for software to get slower not faster.

Indeed, that's what history has shown us. Software usually gets fatter and slower. All the achievements done by the hardware folks is normally negated by the software.

Re:Fairness

[Splab]

I bet you are also the kind of Linux fanatic who keeps telling people, "the source is right there, fix it yourself".

Even though source code is available it can be quite a daunting task to figure out why something is running slow, end even then you are still faced with the task of making it go faster. Poking around some CMS or other (usually) small scale system with a few thousands lines of codes is not a biggie, but trying to fix something with 100.000+ lines of code is pretty much impossible unless you have extensive knowledge of the project.

Also please tone down the Linux zealotry, you are more than likely hurting Linux image rather than achieving what you want.

Re:Fairness

[not already in use]

No, because the OS is closed source you'll always be at the mercy of the license provider to boost performance. Honestly, how much does Microsoft _really_ care about Java performance? Not a hell of a lot.

This is a typical OSS fallacy. "It's open, that means you can fix everything yourself!" Please. 99.9% of people who have an issue with OSS are at the mercy of the maintainer, whose license states that their software comes with no warranty. Much to the contrary, when you pay for a license you have more weight to throw around. Maybe not so much as an individual licensee, but corporations can certainly raise a stink and get issues resolved quickly.

Even the most robust Linux software *is* that way because of support contracts, paid development driven by industry demands.

Re:Fairness

[Vellmont]

This is a typical OSS fallacy. "It's open, that means you can fix everything yourself!"

Ignoring for the moment this discussion was posted under "developers", and therefore relates to the actual people who CAN fix it themselves, I'll respond more generally.

Being at the mercy of "anyone whom you wish to find to fix the problem" (which may or may not include the maintainer) is a lot better than being at the mercy of one entity (the license owner). Most people can't fix their own car either, but would you really prefer a car that you can only go to the dealer to get it fixed?

Much to the contrary, when you pay for a license you have more weight to throw around. Maybe not so much as an individual licensee, but corporations can certainly raise a stink and get issues resolved quickly.

The difference is that with closed source software you HAVE to have a lot of weight to throw around to get it fixed. At least with open source software even an individual has a chance of getting something fixed. Not really the case with closed source. As far as large corps go, they're MUCH better off with open source software, since they can just throw some money at the problem and get it fixed. If they bought the support contract from a 3rd party, even better. If the support contractor doesn't want to fix it, they can ditch that contractor and do it themselves. So you're really trying to tell me that having MORE options, and less of a barrier is a bad thing?

Re:Fairness

[Anonymous Coward]

How many people actually understand the Linux code intimately enough to be able to tell where a problem is? Open source is nice and all but this is not a hello world program. You would need months to study the code and learn it to be able to figure out exactly what is going on under the hood. And you would need to understand several codebases, the VM interacts with the host operating system. Is the problem in the VM itself? Could it be maybe a hidden threading issue? what about a memory leak? Could it be the underlaying system slowing the VM down? How can you tell the difference?

Just because it is open source does not mean all problems are trivial exercises. If that was really the case then every single OSS project should never have any bugs a year after release.

Re:Fairness

[Anne Thwacks]

when you pay for a license you have more weight to throw around

Have you tried reporting problems to MS? Half a day on hold, and then you get cut off. You get a hell of a lot more support for OpenSource (ie e-mail the maintainer, get bug fix back - yeah, I know it may take months, but when did you ever get a Windows bug fixed?).

Re:Fairness

[not already in use]

Ignoring for the moment this discussion was posted under "developers", and therefore relates to the actual people who CAN fix it themselves, I'll respond more generally.

I'm a developer. That doesn't mean I have the skill set to and muck with the java vm. It also doesn't mean I have time to learn and understand any particular code base to make a fix.

Being at the mercy of "anyone whom you wish to find to fix the problem" (which may or may not include the maintainer) is a lot better than being at the mercy of one entity (the license owner).

Sure, in a perfect world where people gladly donate their time to cater to your particular need . This is an empty argument. Just because, hypothetically, someone can fix it, doesn't mean they will, and in my experience over the year, they won't.

The difference is that with closed source software you HAVE to have a lot of weight to throw around to get it fixed.

It's no different in the OSS world. Most OSS worth anything has funded development by people with weight to throw around. Ubuntu linux is popular because someone had a bunch of money and threw their weight around. The benefit to it being open source is that anyone can use it, but without weight to throw around, unless they are a skilled developer with a lot of time on their hands, they have little to no chance that their particular needs will be catered to.

As far as large corps go, they're MUCH better off with open source software, since they can just throw some money at the problem and get it fixed. If they bought the support contract from a 3rd party, even better.

Is that why most large corps are running Microsoft on the desktop and closed source unix implementations on the back end? You can't hold some non-entity accountable when their open source product fails. The whole "no warranty" clause is a pretty big deal when you're big business. Third party support can't cover that.

Re:Fairness

[jo_ham]

Just to latch onto this with someone who's had more experience with it, I just recently installed Ubuntu 8.10 onto my 15" Powerbook to give it a go (1.5Ghz, ATI 64Mb graphics), and it runs much hotter than OS X.

I was expecting something like this, since obviously all the OS X software is tuned up to work with the Powerbook, and that Ubuntu is doing its best but can never be as good [at using the sensors and fans] as OS X without extra help from Apple.

I installed the sensor applet and am getting consistent 50C and 54-55C temps on the cpu and gpu, and the fan seems to have "on" and "off" modes - I am assuming then, that this is down to Ubuntu's less than perfect ability to monitor and control the fans, and also the graphics driver for the ATI chipset not being as optimised as the one provided for OS X.

I am a little lost beyond this, although I guess that's what this experiment is all about, since my previous Linux experience is nil (other than the little bit of Terminal stuff I have been doing in OS X).

Re:Fairness

[Anonymous Coward]

Hahaha! Bug fix back 1/10th of the time, 8/10 times you will get WONT FIX and the other 1/10th 5 months will pass while the maintainers dither about whether its actually worth it.

Re:Fairness

[Anonymous Coward]

So they have been continually improving for 10 years and they still suck?

Re:Fairness

[Yfrwlf]

Much to the contrary, when you pay for a license you have more weight to throw around.

Geezus, first off, why can't everyone here admit that a particular license on software doesn't directly mean that the code under it is either good or bad. You can have shitty OSS and you can have great CSS, you can have great OSS and shitty CSS. It's possible to have any kind of quality of code. Nuff said about that.

Going back to your above quote, that's BS. Just like above, the license a particular piece of software has over it doesn't mean jack as far as problems getting fixed quickly, it completely depends on the amount of support something has. You can pay for CSS support that could be lightning quick with their responses to bug fixes or adding features, and with OSS again, the company you choose to pay for bug fixes, if any company at all, can respond very quickly, and if you know how to fix it yourself you can respond quickly as well. CSS simply makes the latter impossible, but again, their license doesn't mean you will or will not have bad or good support for software under a particular license.

What OS vs. CS software effects is a business' long-term investment and costs. OSS gives you more transparency and less lock-in in general, since you're much more in control of the code, so you can argue that you're much safer from the whims of a particular company. One thing to note is that the costs of CSS were astronomically high, and because of OSS the prices are having to come down to meet the competition. More of that money goes to the developers instead of into exec's pockets. You could say that in theory, eventually things could reach a point at which CSS was as expensive as OSS to improve and maintain, but in reality you have to realize that it being more open is in itself a feature, a feature which obviously CSS will never have, not to mention the freebie fixes and improvements that you have to add on to the paid support which is already there, something which CSS also can never have.

IMO, in the areas where there is still a gap between CSS and OSS, like games for example, will continue to close, and appropriate business models will develop for some of those things and companies will shift at least some of their model to a more open source model. Microsoft has arguably "tried" to start doing that, even though their attempt comes off a lot more as one which tries to disrupt, disorient, and convince the OSS crowd to be more CSS-ish, like developing "open source" software just for Windows, haha. Good one Microsoft, good one, a real barrel-o-laughs on that one.

Re:Fairness

[DragonWriter]

Sorry, that's just a stupid argument. Because an application is closed source they're never going to boost performance?

Uh, no.

The argument was because the product is open source (and implicitly, though this wasn't stated by GP, an open development process) it is easier for outside observers (1) to confirm the underlying root of the performance problem, and (2) to confirm that progress is underway to correct that root cause.

It is not that because the other product is closed source it will necessarily never improve.

Re:Fairness

[not already in use]

I agree with everything you said. The OP's argument was that OSS has inherently good support, while the opposite is true for CSS. In arguing that point, I could see how it would come across that I was saying the opposite, but that is not the case.

Re:Fairness

[Anonymous Coward]

So did the people who performed these Java benchmarks look at the Linux source code to determine either of the things you said?

Re:Fairness

[Yfrwlf]

OK, gotcha. ^^

At the same time though, while it is theoretically possible for them both to be the same as far as quality of support and costs, one would be foolish to call them that once you factor in the real world, i.e. everything together. I believe that in the long run, open source software is more capable of having better support because:

1) More competition to support a particular software. You don't have to rely on a single company for support since any company can, so software companies become simply support and development companies that have expertise in different areas. Less lock in means lower prices.

2) You can support it yourself, or rely on community/free support, if wanted, much much more easily than any closed source software could ever offer.

Of course again, I'm not saying that CSS can't have better support, but the nature of the ecosystem may allow for, more often, better OSS support to exist. At least, in the future, right now I'm not entirely sure how well the development and support companies are doing vs. closed source. The company I work for uses it though and pays some open source companies to develop new features for them. What would be even better is if corporations and organizations could pool their money for the development of common needs, so big features could be paid for more equally. Any way, ya. :P

Re:Fairness

[Vellmont]

I'm a developer. That doesn't mean I have the skill set to and muck with the java vm.

OSS isn't a panacea, and it's not going to give you things you don't have. It just means it's POSSIBLE to fix those problems. With closed source software, all the training, knowledge, and possibly even money in the world won't help you fix the closed source software.

Sure, in a perfect world where people gladly donate their time to cater to your particular need . This is an empty argument. Just because, hypothetically, someone can fix it, doesn't mean they will, and in my experience over the year, they won't.

Who the hell said anything about donating? Either pay for your fixes, or DIY. Nobody "donates" their time to fix my car, why should I expect software to be any different? You seem to be under the mistaken impression that "free" software is all free as in beer. Sorry, but that just ain't the case. Software costs, sorry.

Most OSS worth anything has funded development by people with weight to throw around.

You think? There's plenty of OSS that's funded by people without any money. I've been paid to produce OSS. I'm not sure what you mean by "worth anything", but the software I've made produced value and runs a business. There's tons of other examples of exactly this happening. You don't have to have a million dollars to have someone fix your OSS problem.

unless they are a skilled developer with a lot of time on their hands, they have little to no chance that their particular needs will be catered to.

Well you sure as hell aren't going to be catered to by just asking someone nicely to do it for free. You _could_ pay someone to fix your problem.. developers take money you know.

Is that why most large corps are running Microsoft on the desktop and closed source unix implementations on the back end?

"Most"? I'd say "most" large corps also have some OSS software running somewhere. Why wouldn't they? It'd be stupid to not take advantage of a lower cost, no lock-in option. They don't do it more because the largest corps are just slow to change because they view it as too risky. Stop thinking that OSS claims to solve everyones problems 100% for free.

You can't hold some non-entity accountable when their open source product fails.

There's this company called Red Hat that provides that "responsibility" you're taking about. Maybe you've heard of them? Sheesh, your responses sound like they're from 1997. Perhaps this topic has fallen through a wormhole, and you're replying in 1997? If so, 2008 is a bit different.

Re:Fairness

[abigsmurf]

The only people who are able to sift through hundreds of thousands of lines of code and spot areas to improve on are people incredibly familiar with the code and probably the main development team. It is not easy to understand other people's code, it's even less easy to spot areas that can be optimised out of the blue.

Most improvements come after people have actually run the software, used the drivers, run benchmarks and noticed that the program is buggy or sluggish in certain areas.

Re:Fairness

[DragonWriter]

The only people who are able to sift through hundreds of thousands of lines of code and spot areas to improve on are people incredibly familiar with the code and probably the main development team.

That may be true if one assumes that the method of confirming the problem is a manual searching through the source code.

Most improvements come after people have actually run the software, used the drivers, run benchmarks and noticed that the program is buggy or sluggish in certain areas.

Which is much easier with open-source projects where its much easier to conduct those kinds of experience, whether or not you review any more of the source code than is necessary to understand the interfaces of the components you are replacing.

The benefits of open source are not all dependent on people reviewing existing source code line by line.

Re:Fairness

[Vellmont]

By your argument Linux in theory should be light-years faster then Windows Mac OS or any other OS

Why should linux be faster simply because the source is available? I didn't say linux is faster than windows because the source is open.. I believe what I said was that open source lets the software consumers control their own destiny (or a rough approximation of that).

However if Microsoft could whip Linux in Java Performance they would love it. Fine Vista loss these benchmark but the excusing the one area it did win was in very poor taste.

You seem to have a very strange attitude, as if this is a sports competition with winners and losers. Microsoft really doesn't care about Java performance. Why would they? Java is a cross platform language, and Microsoft has always tried to get everyone over to ONLY the Windows platform. It's really at the heart of the Microsoft strategy.

As far as "excusing" goes, that also seems a strange attitude. This isn't a competition, it's a comparison. As in something that might actually be useful to someone who's interested in performance. I actually _am_ a Java developer, and I don't do anything 3d graphics related whatsoever. So tell me why the 3d performance of either platform is at all relevant to me?

Re:Fairness

[ozphx]

OSS isn't a panacea, and it's not going to give you things you don't have. It just means it's POSSIBLE to fix those problems. With closed source software, all the training, knowledge, and possibly even money in the world won't help you fix the closed source software.

Thats not entirely true. MS will happily take your money to maintain legacy apps - like NT4 if you so choose.

Probably better to pay them as well IMO - its not like you would want to skill up a bunch of in-house guys on MS's codebase :P

Re:Fairness

[QuietObserver]

I'd like to add that with Open Source, making a complaint about a problem with the software is much, much easier. I use ubuntu, and I've had an inexcusably annoying problem with GIMP for some time, and finally got around to reporting it last night. It did take me a few minutes to find the bug reporting tool, but once I did, I was able to issue my complaint in explicit detail, and make sure that what I had to complain about wasn't something that had already been issued. Furthermore, the tool let me know exactly how much information I needed to provide to ensure that my report was as useful for the developers as possible. To my knowledge, such an issue with Microsoft would be next to impossible to achieve without spending hours on the phone with their tech support agents and corroboration from multiple other users.

Re:Fairness

[Anonymous Coward]

Who cares? Drivers are part of the OS so if Linux has crap graphics drivers then it is a failing. Trying to artificially separate components isn't going to fly here.

Re:Fairness

[AngelofDeath-02]

I'm not sure why this was marked as a troll, because it's fairly accurate. Various companies I've worked for have requested customized versions of software to fit their needs and performance demands; they are big enough to do it. While that company might be big enough to throw money at licenses, they may not have the talent to reprogram an open source project to accomplish the same thing.

Re:Fairness

[Anonymous Coward]

You'd need to be able to examine both sets of drivers in order to determine realitive badness. So your argument is bunk.

3D in Java?

[SilentChris]

The two OSs were similar in ray-tracing, and Vista was faster at Java OpenGL due to shortcomings with the Linux graphics driver.

I know this will be seen as a troll, but who the hell uses Java for ray-tracing or with OpenGL?

Re:3D in Java?

[Nicopa]

I think you should check the Java port of Quake then: http://bytonic.de/html/jake2.html

Re:3D in Java?

[gzipped_tar]

It's just a model. Oh, I mean "test".

Re:3D in Java?

[pak9rabid]

I know this will be seen as a troll, but who the hell uses Java for ray-tracing or with OpenGL?

I'm not sure about the former, but there have been some interesting uses of the latter. Check out Jake2 sometime. It's Quake2 recoded in Java with full OpenGL support. Obviously it's not going to perform as well as it's native counterpart, but it is very playable and cross-platform.

Re:3D in Java?

[Anonymous Coward]

My understanding is that there's a small movement towards using java in fact for 3d stuff. Using Java makes it easier to port cross-platform and all the 3d stuff occurs on the video card anyways so there's *practically* no java overhead. Of course, this is info I read out of a book a long time ago and I don't really keep up with that stuff so that 'movement' could be dead now.

Re:3D in Java?

[windsurfer619]

Direct link to webstart online deathmatch

Re:3D in Java?

[Blakey Rat]

Grandparent asked who USED it, not "who created pointless, idiotic demos of it".

If you bought a copy of Quake, would you get the Java one? No? Then nobody uses it.

Re:3D in Java?

[Anonymous Coward]

I ride yo momma like a fucking circus pony.

Re:3D in Java?

[Anonymous Coward]

To further this comment, who in the Hell uses a Quiche-Eater language like Java?

Re:3D in Java?

[Gastrobot]

Raytracing with Java - http://www.artofillusion.org/

Not completely fair

[Yahma]

While interesting, the results pose more questions than they answer. I can see several problems with the benchmark:

• Slightly different Java versions used in the tests.
• The tests should have all been forced to use either Client or Server mode.

Its possible that the Windows Java defaults to client, while the Server mode was used in Linux. That could account for the major speed improvements seen in the Linux versions. I would like to see the Server mode forced on all the JVM's and the tests re-run.

Re:Not completely fair

[TobiasS]

all jvms except for the 32 bit win32 version default to -server for this type of hardware.
server class hardware for the jvm is defined as min. 2 cpu's and 2GB memory both of which are present in the test machine.

I saw no mention in the article whether the vista install was 32 or 64 bit.

If it was running with default vm options it could make quite a difference on vista.

Re:Not completely fair

[ozphx]

The "slightly difference" version was a fairly major point release too (evidently especially from a performance perspective).

I mean seriously - benchmark *nix vs doze on things like fileserving / web serving / TCO / etc. Execution of native code? Please..

Not surprising

[greg_barton]

I've been using scimark for years to evaluate system performance with java.

Try it yourself.

Linux has outperformed windows (on average) for years, and OSX as well until recently. (java 1.4 performance on OSX was dismal)

Re:Not surprising

[owlstead]

And Solaris. But I would guess that Solaris is pretty much like Linux in many expects (or Linux like Solaris, if you will).

Re:Not surprising

[owlstead]

Whoops, that's what you get when you use this new interface. That was the wrong thread entirely, don't know what happened. Anyway, it has already been said in the "I was surprised" thread.

32 vs 64 bit

[tarunbk]

Any equally good benchmarks, which also include a comparison of the 64bit and 32bit java versions... now that java has a 64bit browser plugin, i don't have to worry about separate browser and stuff... just waiting for the final version of 64bit flash to come out...

Re:32 vs 64 bit

[tarunbk]

will there be near future when we have to worry about 64 vs 128bit ?? or will cpu instructions also go the core way, and have as many multiple of 32 as one wishes? never know the future...

Re:32 vs 64 bit

[Anonymous Coward]

To my abysmal surprise, the unstable beta of Adobe's Flash64 plugin hasn't crashed my browser yet. I am running the same build of firefox, just a 64bit one on my Fedora 10 x86_64 laptop. The old flash32 on a 32 bit system crashed like there was no tomorrow. This time around, I have yet to experience a crash. What changed, I ask?

I was surprised

[spottedkangaroo]

To no-one's surprise, Ubuntu was faster

I'm surprised.

I'd expect Java to go faster in windows. I don't think the reasons for using Linux are speed and software availability. I'd expect much more attention is paid to the windows versions.

Re:I was surprised

[nargoes]

I wasn't surprised. We have developers using the same versions of build tools, JVMs and hardware. A build on Ubuntu takes about 4 minutes. On a (freshly imaged) windows box the same build takes at least 10 minutes. Very empirical maybe, unsurprising none the less.

Re:I was surprised

[Brad_McBad]

Really? Why don't you think speed is important to servers running linux?

Last time I checked there were plenty of linux servers running high powered linux apps.

Re:I was surprised

[Brad_McBad]

Sorry, that should be "java apps"

Re:I was surprised

[Anonymous Coward]

You were correct anyway.

Re:I was surprised

[Brad_McBad]

Meh, it's Friday afternoon. I'm allowed a couple of screwups, surely?

Re:I was surprised

[windsurfer619]

Well of course you weren't surprised now because you already knew, but I bet you were surprised when you saw it for the first time.

Re:I was surprised

[LWATCDR]

You shouldn't be.
Most java development these days takes place on the server side. Linux is has a large precentage of the server market. Then you must know that Sun is a Unix company. They push Solaris and java on Solaris. Solaris is a lot more like Linux than Windows. The the final piece is that in the Windows server market Java shares space with .net.
So as far as the amount of attention I would say that Solaris/Linux/Unix gets just as much attention as Windows does.

Re:I was surprised

[ZerdZerd]

I'm also surprised. Every java-program I've run on both windows and Linux was slower on Linux. For instance I made a program which fetched the contents of the clipboard. On windows it took 33ms, and on Linux about 800ms.

I also discovered a lot of inconsistencies in "cross-platform" java. When I push and hold down a button in windows it's received as Down - pause - down down down... And on Linux it was for some reason Down Up Down Up...

Re:I was surprised

[Shados]

Don't be surprised. The benchmark is comparing a JVM version that recently got some of the biggest performance improvements in Java's history, vs the version that preceeded it... You'd get similar number if you did the same and compared Windows vs Windows or Linux vs Linux.

Re:I was surprised

[shutdown -p now]

I'm also surprised, but for another reason. A few years ago, it was common knowledge that Sun optimizes Java for Solaris first (duh), Windows second (Sun has a long history of "partnership" with MS), and Linux last (why support competition for the server market?). Either things have changed, or the study isn't all that good. Seeing how they compare two different Java releases, I guess it's the latter.

Re:I was surprised

[Fujisawa Sensei]

But Microsoft spends a huge amount of effort making sure that Java is a slow as possible on Windows machines.

Take Hero Designer, takes a few moments to start on both machines. But leave it running, and then go out to lunch. On Linux its there waiting for you, on windows you're better off rebooting, than you are trying to switch back to it.

Re:I was surprised

[shutdown -p now]

But Microsoft spends a huge amount of effort making sure that Java is a slow as possible on Windows machines.

Do you have links to prove it?

Do not assume (Microsoft's) malice where (Sun's) incompetence is enough to explain things.

By the way, why would Microsoft deliberately make Java run slow under Windows? It makes no marketing sense - it just means that whoever wants Java (which is a lot of people) would just use a Solaris or Linux server instead - and that isn't good for MS. They'd rather have you run Java on Windows Server and pay them for the license (bonus points if you also use MSSQL, which is my Microsoft actually provides JDBC drivers for SQL Server).

Re:I was surprised

[Anonymous Coward]

Actually I was somewhat surprised. I use Java on Linux because it's been easier to manage. I'm hardly a Java guru, but have spending lots of time configuring tomcat and JBoss recently, plus spending immense amounts of time in front of Eclipse. It's been far easier for us to create a ksh script to select a JAVAHOME on Linux than it has been to maintain multiple VMWare Windows installations to guarantee we're using a given version of Java. Though we do a cursory test on Windows, our targets are on Linux. Given the size of the Windows market, I'd made the assumption that Linux (like all too often) was a second-class citizen.

So I'm surprised... :D

Re:I was surprised

[Fujisawa Sensei]

But Microsoft spends a huge amount of effort making sure that Java is a slow as possible on Windows machines.

Do you have links to prove it?

Do not assume (Microsoft's) malice where (Sun's) incompetence is enough to explain things.

Gee I don't know, how about the non-public APIs they use, vs the public APIs they give everybody else? Was even documented in the MFC/C++ manual, that Microsoft did not use that API for office.

You'll also notice that while they made C# an open standard, they did not make the standard API part of they.

By the way, why would Microsoft deliberately make Java run slow under Windows? It makes no marketing sense - it just means that whoever wants Java (which is a lot of people) would just use a Solaris or Linux server instead - and that isn't good for MS. They'd rather have you run Java on Windows Server and pay them for the license (bonus points if you also use MSSQL, which is my Microsoft actually provides JDBC drivers for SQL Server).

How about because they want to lock you into their product suite with C# and asp?

Microsoft does not play nicely with others, ever. Even when MS is being nice, never trust them, they're like a crack dealer giving out free samples.

Having issues with performance? Don't change your OS, change your apps.

Re:I was surprised

[bckrispi]

I'm also surprised. Every java-program I've run on both windows and Linux was slower on Linux. For instance I made a program which fetched the contents of the clipboard. On windows it took 33ms, and on Linux about 800ms.

On Linux, do you tell the JVM to run in '-client' mode? Linux JVM's, unlike those on Windows, default to run as '-server'. This can lead to a less-responsive feeling GUI application.

Re:I was surprised

[shutdown -p now]

Gee I don't know, how about the non-public APIs they use, vs the public APIs they give everybody else? Was even documented in the MFC/C++ manual, that Microsoft did not use that API for office.

Erm, originally you said that Microsoft somehow makes Java slower. What you describe is a way to make MS's own products faster. There is a difference here, most certainly in the context of this study (which is Java/Win vs Java/Linux, not Java vs .NET).

You'll also notice that while they made C# an open standard, they did not make the standard API part of they.

Actually, they did make some of the APIs part of the standard (it's part of the CLR standard, not part of the C# standard) - what they call the Base Class Library, BCL. This includes System, System.Collections.*, System.Text, System.IO, System.Xml, and quite possibly a few other things I can't remember right now.

What's proprietary is System.EnterpriseServices, WinForms, ASP.NET, ADO.NET, WPF, WCF and WWF. Which isn't that good, but still, on the whole, the ISO-standardized CLR APIs still cover way more than e.g. the ANSI C++ standard library.

How about because they want to lock you into their product suite with C# and asp?

Crippling Java on Windows alone doesn't serve this purpose - in this case, Java still remains competitive with .NET performance-wise, so long as you use a different OS (and then MS loses money from OS sales as well!). If they could somehow do it across all platforms, or do an "embrace & extend" tactic on Java/Win, that would make sense - and they actually tried that with J++, but Sun sued them back then and won, so that won't fly again.

At the moment, Microsoft tactics seem to be to try to make everything work on Windows - e.g. they work directly with PHP team to maximize performance of PHP on IIS, they produce and maintain open source MSSQL drivers for PHP, and then those JDBC drivers I've mentioned already. This also makes sense - if you run Java or PHP on Windows and not Linux, it's still a net win for Microsoft, and certainly preferable to you running Java or PHP on Linux. And to make sure you'll want to run it on Windows, they have to make it run better in some ways (faster, more convenient... whatever).

It's all really just basic business sense. MS is not out there doing evil for the sake of evil, they're just making money in any ways they can. If selling Windows for Java servers pays, they will sell it (just like Sun will sell Linux servers on its hardware, even though they'd prefer you to buy Solaris instead - but they won't let a customer go away because of that).

Re:I was surprised

[intheshelter]

I'm surprised anyone cares about this test. One OS is bloated, overpriced and useless, the other is free. . . and useless. Based on the news I've been hearing over the last year it doesn't appear the public is particularly excited about using either one.

Re:I was surprised

[Shados]

If thats true, then we know right there why the benchmarks were favoring Linux... -server is superior in basically every ways, shape and form for synthetic benchmarks, since -client is for feeling of responsiveness, not raw performance, and the article said nothing about forcing the VMs to run in the same mode in both cases... not very fair to compare benchmarks not using the same mode...

Re:I was surprised

[Fujisawa Sensei]

Erm, originally you said that Microsoft somehow makes Java slower. What you describe is a way to make MS's own products faster. There is a difference here, most certainly in the context of this study (which is Java/Win vs Java/Linux, not Java vs .NET).

When you have control over the application, and the APIs there's no difference, especially when you can train the OS to recognize various kinds of binary files.

It's all really just basic business sense. MS is not out there doing evil for the sake of evil, they're just making money in any ways they can. If selling Windows for Java servers pays, they will sell it (just like Sun will sell Linux servers on its hardware, even though they'd prefer you to buy Solaris instead - but they won't let a customer go away because of that).

They're the business practices of an convicted monopolist. They support other products to draw support various customers, then they snuff the support, and migrate the customers to their own proprietary platforms.

Re:I was surprised

[Anonymous Coward]

I knew java on linux was faster before i saw these benchmark results. just run the same java app side by side on the same hardware one on windows the other on linux and notice the staggering difference. it's the one thing that impresses my colleagues always the most when using linux (which is sad in a way).

Vista beaten by other OS "to no-one's surprise"

[markdowling]

If it's not surprising, it shouldn't be on the front page, no? Doesn't it fail the "NEWS" for nerds test?

Vista beats anybody would be in the same league as man bites dog and very news-worthy.

Re:Vista beaten by other OS "to no-one's surprise"

[moderatorrater]

News can simply be notifications of an event, in this case an event that gives us Ubuntu users another reason to gloat. I'm having dinner with a windows sysadmin tonight, so I suppose I owe Slashdot some thanks for giving me a good topic of conversation.

Re:fast java

[Anonymous Coward]

Clearly, this is a troll; I cannot help but reply to it.

When was the last time you actually wrote any software in Java? Describing the Java of today based on your memories of the old Java is like comparing Linux today to Linux of the late 90's. There have been several architectural changes both behind the scenes and in front of them for both.

I'm sure it was a fair test...

[Anonymous Coward]

Does the Windows Java implementation use Windows native calls everywhere or was it written to Linux/Unix and using some layers to map that to Windows native?

Bah, theoretical Java performance

[netsavior]

So javaQuake in an applet viewer is faster, try using ANY browser available for Ubuntu and load the applet in it... now use a naked IE activeX control (or even an IE window) to host the applet.

Linux fails, because the overhead of running ubuntu gets COMBINED with the overhead of running FireFox of Epiphany or whatever, whereas windows overhead already includes IE.

Pretty much the only thing "taxing" I use my laptop for is playing Runescape (browser hosted 3d rendered 100% java applet MMORPG) and with the same laptop dual-booting WinXP and Ubuntu, it is 100% flawless under windows in IE and almost unplayable in Ubuntu for any combination of browser and JDK that I have tried. I am almost to the point of writing my own browser applet container (the game requires a browser, 100% applet container doesn't work), it is just really annoying to switch to "windows the hog" just to play a simple video game that SHOULD be cross platform (and is as long as the linux system is higher spec than my laptop)

Re:Bah, theoretical Java performance

[H0p313ss]

Pretty much the only thing "taxing" I use my laptop for is playing Runescape

Please turn in your geek card now.

Re:Bah, theoretical Java performance

[netsavior]

A laptop dedicated to an weird MMO is not geeky?

I didn't say it was the most taxing thing I do on computers, just on that particular laptop...

Re:Bah, theoretical Java performance

[NullProg]

it is 100% flawless under windows in IE and almost unplayable in Ubuntu for any combination of browser and JDK that I have tried

Your doing something wrong. What I dont know (Make sure your using the SUN JVM instead of the default OpenJDK). My twelve year old plays the premium Runescape every day under Ubuntu+Firefox, no problems and its much faster than the windows machine.

Enjoy,

Re:Bah, theoretical Java performance

[Just Some Guy]

Linux fails, because the overhead of running ubuntu gets COMBINED with the overhead of running FireFox of Epiphany or whatever, whereas windows overhead already includes IE.

Yes, because IE is compiled with -fomit-instructions and runs infinitely fast.

Re:Bah, theoretical Java performance

[H0p313ss]

A laptop dedicated to an weird MMO is not geeky?

Nethack would be geeky, Runescape is just wrong...

N.B. I'm mostly kidding.

Re:Bah, theoretical Java performance

[Anonymous Coward]

This is probably an issue with your graphics card driver, not Ubuntu itself.

Re:Bah, theoretical Java performance

[Anonymous Coward]

No, it's just stupid.

Re:Bah, theoretical Java performance

[nxtw]

Linux fails, because the overhead of running ubuntu gets COMBINED with the overhead of running FireFox of Epiphany or whatever, whereas windows overhead already includes IE.

Nonsense. Windows overhead doesn't "include" IE; the shell uses the IE rendering engine. If you run Firefox in Linux, you've got one copy of the Firefox rendering engine loaded and you're using that copy of it in Firefox. If you run IE, you've got one copy of rendering engine loaded and both the shell (explorer.exe) and Internet Explorer (iexplore.exe) use it.

The only potential advantage here is that the rendering engine might already be in physical memory when you start IE.

I suppose you could possibly "waste" some memory in the Linux scenario if the UI had loaded a HTML rendering engine other than the one your browser used. But it would just get paged out if you ran out of memory...

Re:Bah, theoretical Java performance

[Fujisawa Sensei]

This is probably an issue with your graphics card driver, not Ubuntu itself.

Since the graphics driver is running as part of the kernel, it is part of Ubuntu and perfectly legitimate to include in a benchmark.

Re:Bah, theoretical Java performance

[Lennie]

I know from my non-geek nephew that qemu with Windows on Linux can run Runescape just fine.

Re:Bah, theoretical Java performance

[Anonymous Coward]

For me one of the worst thinks about linux graphics drivers are that their default settings are horrible performace wise. I have Ati mobility radeon 9200 (which is rv280 card) and when I run it with default settings glxgears gives me 1750-1800 fps but when I tweak xorg and dri settings to the best stable settup I get 2600-2650fps and this isn't all that hw could do. I have even seend over 3k fps in glxgears but then my system ussualy hard locks in about 15 minutes. But this doesn't help java. It seems like java graphics implementation inlinux doesn't use my hw accerlation correctly.

btw, just a note: I can get now better performace from HL1 under wine than I had 5 years ago in win xp. But problem 5 years ago was that I didn't know how to tweak win xp driver same way like I can my linux driver now.

so what to do:
1. Enable EXA in xorg.conf
2. Enable all others setttings that promise minor speed boost one by one and allways test a opengl game (or compiz) for an half hour if that setting is stable in your computer.
3. use driconf to gain the last boost for 3D.

Resources available to JVM

[tukang]

It's possible that Java is actually better implemented for Vista but because Vista uses more resources it ends up running slower. It would be interesting to see a comparison where the amount of available resources is equal.

But do all applets work?

[armer]

Sure it may be faster, but do all applets work properly? For the life of me, I cannot get the applets from Pogo.com to work under ubuntu (fresh install of 8.10 with restricted extras). Xp and Vista work great. Its good to be fast, but it needs to work too...

Re:But do all applets work?

[Anonymous Coward]

I have had them working under Fedora 8. I installed java from Sun and followed their directions on their website to install it.

I think pogo.com might have required the Microsoft Virtual Machine at one time. However, it seems to work now.

Re:But do all applets work?

[armer]

It did, but that was more than a year ago... I guess with the new version (.10 java) I will have to fire up virtual box and try again...

Java on desktop has been irrelevant for a while

[BigGerman]

Now, if they tested webapps performance, say on Tomcat on both OSes, the results would be quite different. I would guess Vista would be faster for a single thread test but would not scale at all.

Re:Java on desktop has been irrelevant for a while

[Shados]

If you wanted to do that, then you'd want to test on WinServer 2008, not Vista, in which case it definately WOULD scale.

Re:Java on desktop has been irrelevant for a while

[plague3106]

Well, if that's the test you're after, why wouldn't you test a server version of Linux against Windows 2008?

Re:Java on desktop has been irrelevant for a while

[PitaBred]

Dunno about that. We have a tomcat app as our main product, and we've found invariably that Linux performs better than Windows with it.

Re:Java on desktop has been irrelevant for a while

[Jimmy King]

Java on desktop is irrelevant in what way? As in the performance is good enough that it doesn't matter or as in it's not really used on the desktop, anyway? I'm curious because I have several java desktop apps that I use on a regular basis in both windows and linux.

Re:Interesting, but lacking some crucial details..

[xouumalperxe]

Indeed, you're rambling. Partitioning schemes are not very relevant for this test suite. The only test that could realistically be considered HDD-bound is the file encryption one, and even then the performance difference is too large to be attributed to which bit of the disk you're reading. I'd guess that on-the-fly encryption is must more CPU-bound than it is HDD-bound, and under that light that test proves consistent with all the other raw maths performance tests.

If anything, what I'd like to see is the 3D rendering bit redone with an nVidea card (that actually has decent drivers), to actually test their assertion that the performance loss on the linux side is really due to the graphics drivers.

Re:Interesting, but lacking some crucial details..

[chrb]

Various problems with the Phoronix test methodology have been noted before and before that. Without going over the same stuff, here are some potential questions about this benchmarking:

• Where is the statistical analysis of these results - ok, you ran a test once and it was 30% slower. Is this reproducible? What is the variance? Is there any statistical difference between openjdk/sun java?
• Why is the Java minor version different? Do you see the same results if the same minor version is used?
• As mentioned in the previous discussions, exactly why is Windows slower on the file encryption task - it should be either limited by disk throughput, or by CPU throughput, so observing a 40% drop in performance attributed to the underlying I/O handling of the operating system is somewhat surprising; are you sure the test methodology is sound here, and if so, how do you explain the results?
• Are these results applicable to both 32 and 64 bit distributions and JDKs?
• How do you know that the 2D benchmark performance on Linux is attributable to poor graphics drivers? Why not run the test on another PC and then swap out graphics cards (hence eliminating all other factors) and report on the results?

There are a lot of questions that this benchmarking should have answered, and a lot of assumptions made that could potentially be invalid.

Re:Interesting, but lacking some crucial details..

[chrb]

even then the performance difference is too large to be attributed to which bit of the disk you're reading

No it isn't. How do you know that a particular sector of the hard disk isn't failing, causing access to that one sector to be a thousand times slower than other sectors? This is why experiments are supposed to be run many times, across different platforms, and the aggregate results taken. Without multiple experimental replicates you have no way of showing that the results you observe generalise at all; the observed problem could just be one bad run.

Re:Interesting, but lacking some crucial details..

[plague3106]

Maybe I'm missing something, but is the code for the JVM itself 100% identical on both platforms?

Also, what about running services? Linux and Windows both have a different set of multiple running services. Yes, if Windows is "doing more" by default than Linux that will slow the application, and you can argue about whehter or not Windows should be doing more, but its not as if the sole purpose of either OS is to run Java applications as quickly as possible.

So... all this tells us is that in their configurations, Linux runs their Java programs faster.. which if thats ALL you care about is fine... but most users also care about other services either OS provides.

Re:Interesting, but lacking some crucial details..

[LWATCDR]

One also has to wonder how well they "tuned" the Vista install.
Was it running antivirus or indexing in the background?
Vista tends to have a lot of "stuff" running in the background. A lot of of which you can turn off.
They also really should have used the .10 release. It was a huge point release.

Ever seen Sunflow?

[nobodyman]

I can't gauge it's performance relative to a c-based raytracer, but you should really checkout Sunflow.

It can produce gorgeous results, and it'll use as many cpu's as you can throw at it. I'm not sure how much commercial use it enjoys, but it's well beyond the "ooh check out this glass sphere" java raytracer applets you typlically see.

Java on Vista...

[Colin Smith]

Why would you want to do that to yourself?

 

Re:Java on Vista...

[Billly Gates]

Need MS Office. Also Netbeans is ugly as hell on Linux but the windows version looks native and even has cleartype fonts

But What About Solaris?

[John Hasler]

After all, that's what everyone is going to be running on their laptops from now on, right?

Re:Interesting, but lacking some crucial details..

[Anonymous Coward]

Also most drives these days have same transfer speed on the inner tracks as the outer, they use the extra space for hiding bad sectors etc.

Re:Interesting, but lacking some crucial details..

[Anonymous Coward]

It is if you use it to read and write files.

Re:Interesting, but lacking some crucial details..

[Anonymous Coward]

yes, because bad sectors can only happen on the outermost rings. Retard.

Re:Interesting, but lacking some crucial details..

[xouumalperxe]

So... all this tells us is that in their configurations, Linux runs their Java programs faster.. which if thats ALL you care about is fine... but most users also care about other services either OS provides.

Precisely. There are so many much larger things at hand at so many levels, that disk layout, bad sectors, etc are really a minor, minor issue.

Re:Interesting, but lacking some crucial details..

[xouumalperxe]

Ooops. Forgot to mention that, also, just because the code is the same doesn't mean that the compiler is, or that the compiler's output for different platforms is, or that the problem isn't in the platform-specific code. Or, in this case and as was mentioned elsewhere, perhaps the dominating issue is that the JVM versions are different and the one used in Linux is head and shoulders above in overall performance.

Re:Interesting, but lacking some crucial details..

[xouumalperxe]

No it isn't. How do you know that a particular sector of the hard disk isn't failing, causing access to that one sector to be a thousand times slower than other sectors? This is why experiments are supposed to be run many times, across different platforms, and the aggregate results taken. Without multiple experimental replicates you have no way of showing that the results you observe generalise at all; the observed problem could just be one bad run.

True, reducing statistical bias is an important point. So is taking all the data into account and making sure your conclusions are consistent with all of it -- and doing so while not adding superfluous factors to the equation.

My argument came in two points. First, that the performance difference was too big to be attributed to disk layout (which you then countered with faulty sectors, which is grabbing for straws). Moreover, I stated that the result was consistent with the other tests. Which is likelier: "there's a hard disk defect on the specific bit of the platter that windows was accessing, and that particular test performed worse than it should, fortuitously yielding a performance difference in line with the other tests" or "this one individual test performed the same as most of the other tests (OpenGL being the exception, for known reasons), indicating consistency between test results"?

Re:Interesting, but lacking some crucial details..

[Anonymous Coward]

i think the more important question is who the hell uses java on a computer anymore?!

Sunflow also good at multi-threading

[nobodyman]

Sunflow multi-threads quite well, but I've never compared it to anything else. Did you you have a chance in your testing to see how well it scaled vs. the c++ version as the number of cpu's goes up?

Re:Sunflow also good at multi-threading

[Yetihehe]

No. I tested my own simple raytracer (started in C++, tried how fast it was in java then moved to java because it's a lot easier and speed penalty is not so bad).

yo dawg

[Anonymous Coward]

so you can virtualize while you virtualize

Re:yo dawg

[Anonymous Coward]

yo dawg i herd yo like virtualization so we put vm into yo vm so u can virtualize while u virtualize

A couple of test vs. scientific benchmark

[DrYak]

Various problems with the Phoronix test methodology have been noted

From the very first paragraph of TFA (Yes, I know this is /. and nobody RTFA, but still) :

Have you ever wondered on what operating system Java works the best? While by no means is it a conclusive multi-platform comparison, for this article we ran a number of Java benchmarks on both Windows Vista Premium and Ubuntu Linux to see how the Java Virtual Machine performance differs.

They are just doing a quick test. Nobody is trying to do a scientifically perfect test.

Now, that aside.

Are these results applicable to both 32 and 64 bit distributions and JDKs?

You're joking, I hope ?~ 64bit Windows ? Are there at enough stable drivers for the thing to be able to at least boot ?~

People have been complaining about insuffisent driver support under Linux, but driver support for 64bits Windows - at least in my experience - is somewhere between catastrophic (Vista 64) and completely inexistant (XP 64 for AMD64).

More seriously :
Most of the "serious" benchmark websites tend to run all benchmarks in 32bits Windows only. Including benchmarks of server oriented hardware (Opteron & Xeon are often pitted one against another running under some 32bit version of Windows).
Not only do I find this a bad practice because it's not representative of the reality out there (most of the server I encounter every day run 64bits Unices - very often 64bits Linux), but now we have a small actual proof that typically server-applications don't run quite the same under Linux and under Windows, thus benchmarking the hardware under both environment might be relevant for all readers running their servers under Linux.
The sad truth is that most "professional" benchmarkers are more used to test high-end enthusiast gamer hardware, which makes more sense to be tested under 32bits Windows. Thus they probably lack the experience. As bad as it is, Phoronix' test suite is a first step toward helping benchmarks under the penguin OS.

Re:A couple of test vs. scientific benchmark

[Tawnos]

When was the last time you tried running Windows 64 bit? My experience is that both Vista 64 and Win7 64 both have great driver support now (never tried XP64, but the XPDM (XP driver model) is deprecated, so I understand why finding hardware for the platform is hard). In fact, the issues at Vista launch were largely due to moving to the new driver model, and hardware vendors not being up to speed. Two years later, those issues are no longer a problem.

Or was your comment sufficiently advanced satire that I'm missing? :)

Re:A couple of test vs. scientific benchmark

[Anonymous Coward]

Are you a retard or just ignorant? Are you living in 1998 or what? There are almost no problems with drivers on 64 bit windows. And most driver issues don't even cause major problems.
Furthermore you are implying Vista is server operating system. For that they should compare windows 2008 server (x64) with linux. win2k8 don't have DRM like vista and most resource hogging services are disabled by default. I am sure result would be much different with win2k8.

Re:A couple of test vs. scientific benchmark

[Bert64]

Driver support for vista is reasonable, if you are using hardware that came with it...
Running 64bit vista on older 64bit hardware, or on machines which were shipped with 32bit vista can often be a significant hassle.
64bit xp is just laughable when it comes to driver support, but hardly anyone ever used it...
64bit win2k3/2k8 tends to have better support, since 64bit has been common in the server space for longer, and most servers shipped with windows are designed to run 64bit these days.

Also, why wouldnt high end gaming hardware be tested under 64bit? You can get videocards with 4gb of ram now which would be completely useless on a 32bit windows, arent these cards designed for gaming?

But i would like to see more 64bit benchmarks, especially linux ones, 64bit linux is a lot more mature than 64bit windows and is pretty much a no brainer if you have 64bit hardware.

Re:A couple of test vs. scientific benchmark

[godefroi]

I've run Windows Server 2003 (64-bit) and XP Pro (64-bit) and I didn't have a single driver problem. Not one.

Re:A couple of test vs. scientific benchmark

[Anonymous Coward]

I believe that XP-64 and 2k3 shared a common code base.

Re:A couple of test vs. scientific benchmark

[setagllib]

4GB video RAM is not an issue on 32 bit because the only part of it the host system can address directly is the "aperture". This is the same as how you can use a 1TB disk even with a 32-bit OS and only 1GB RAM. They don't have to all line up directly. Unless I'm just taking modern operating system design for granted and Windows really does require a direct memory map from all video memory to system memory.

Stock configurations

[DrYak]

One also has to wonder how well they "tuned" the Vista install.

Everything left to default, including desktop effects according to TFA.

You're making a huge assumption here...

[Anonymous Coward]

...namely, that people wish to maintain intellectual honesty. Admitting that Vista is faster than Ubuntu has the potential to cause large amounts of cognitive dissonance.

Re:Interesting, but lacking some crucial details..

[cyphercell]

http://www.cs.berkeley.edu/~kubitron/courses/cs252-F00/lectures/lec20-disks.pdf

I was just having this conversation the other day. According to the pdf linked above bandwidth on the outer tracks is 1.7x times that of the bandwidth on the inner tracks (I assume this variation is greater on larger drives). Definitely worth looking into, but an optimized benchmark is ridiculous if you are testing for desktop use. Better to test a huge amount of default installs.

poor benchmarking effort

[buddyglass]

1. On Windows they use 1.6_07, while on Linux they use 1.6_10. With different JVMs. So we have different JVMs with potentially different levels of optimization, and potentially different supporting JARs.
2. They use Windows Vista SP1, which is a desktop OS version. Ubuntu is not. They should have tested on Windows Server 2008 SP1 or SP2-Beta.
3. They use a graphics-intensive benchmark like OpenGL. Duh. No wonder Vista out-performs: you're testing the video driver, not the Java code or the OS internals.
4. They use a benchmark that heavily reads and writes to disk. However, they don't say whether the Windows and Ubuntu installations are located on separate (and identical) hard drives, or dual-booted on the same drive. If the latter, then one of them is potentially going to have a higher max. transfer rate.
5. We're given no indication as to the JVM settings used for either test. Even if they used default settings, the defaults may be set differently on the Windows and Ubuntu installations. JVM settings can drastically change the way Java code performs.

Re:poor benchmarking effort

[Bert64]

1, agreed... were they using the latest versions available for the respective platforms? i don't see why the windows version should be older, and it has an auto update so it should have tried to update itself anyway.
2, i assume they used the desktop version of ubuntu since they mentioned desktop effects, ubuntu server does not have a desktop environment by default.
3, they should really have tried some other kind of video benchmark so as to clarify the differences in the drivers... i always thought intel video had good linux drivers, but maybe not, a fairer test would have used nvidia drivers which perform about the same across platforms, but this does highlight a problem with the intel drivers for linux.
4, agreed, they should use the same disk and reformat it for the second set of benchmarks really...
5, default settings are perfectly reasonable, since they will be what most people use, and it's easy to verify what the default settings are so no need to manually enumerate them.

Re:poor benchmarking effort

[buddyglass]

2. Even the desktop version of Ubuntu is still a server-class OS, since it's running Linux under the covers. Unless they tweak a bunch of kernal tunables in such a way as to mess with its ability to act as a server. So, the comparable version of Windows to use would be 2008 Server.

3. IMO they shouldn't have used graphical benchmarks at all, since those are going to be drastically affected by driver quality. They're not so much comparing "Java on Ubuntu" to "Java on Windows" as they are "Intel drivers on Ubuntu" to "Intel drivers on Windows".

5. Off hand, I don't know the default JVM settings for Windows and Ubuntu. If they got Java from an Ubuntu repository, it's possible that whoever created the package tweaked the default values. Still, though, those values are configurable. So, if the goal is to test "platform vs. platform", then they should strive to equalize all the other variables, such as JVM settings.

Re:poor benchmarking effort

[Shados]

1 -> I don't know when they ran the benchmark, but when the article was published, even on Linux they were using one update behind. They say in the article they were using the latest...and I don't know whats Sun's release schedule and if they release for Linux first, but I -heavily- doubt the Windows version was 3 release behind... and revision 10 was a very, very important revision in term of performance, so even if it wasn't released on Windows, it was a very dubious time to benchmark things...right after one of the fastest JVM revision came out for one of the OS but not the other (though I doubt that was the case)

Re:poor benchmarking effort

[Bert64]

2, even the "server" version of windows is still a desktop-class OS... they were comparing 2 versions which are marketed as suitable for use on desktops....

3, yes they should have done graphical benchmarks, but they should have done other non java related graphical benchmarks to determine what difference the drivers made, and chosen a videocard for which driver performance is similar, such as nvidia cards... on the other hand, drivers are a significant part of the overall experience so lousy drivers are a serious problem.

5, perhaps, but testing the most common configuration users will be using is still a perfectly valid and useful benchmark.

They ignored a "minor" u10 feature..

[Anonymous Coward]

Java SE6u10 was a major release and one of its most important improvements was... performance. You can read about that here, one of the many blogs. So excuse me for not being the slightest bit surprised that u10 outperforms u7.

Re:They ignored a "minor" u10 feature..

[Bert64]

Strange, windows doesnt come with java by default so they must have explicitly downloaded and installed it, why would they use an old version?
Also java has an update process, would this not try to install the latest version?
Is the u10 update not available for windows?

Java benchmarks are good, but

[strangepics]

Would be nice to see bulk.fefe.de type benchmarks for webserving on that hardware as well, to see how the OSes scale in web serving. I wouldn't be surprised to see either Linux or FreeBSD come out at the top.

Simply Put

[molex333]

Simply put, these results can be attributed directly to the fact that Windows M1cr0sux0r's!!!!

Re:Interesting, but lacking some crucial details..

[Anonymous Coward]

'but its not as if the sole purpose of either OS is to run Java applications as quickly as possible'

But they are general purpose OS and the point of THIS benchmark is to let you know which will give better java performance in its default state.

Which services to install and run by default is just as much part of the design of the system as anything done with the code so leaving defaults is quite fair. It isn't as if you are going to turn all windows services before running open office.

Re:Interesting, but lacking some crucial details..

[shaitand]

'Vista tends to have a lot of "stuff" running in the background. A lot of of which you can turn off.'

That is a design decision of the operating system. An optimized system is NOT representative of average vista performance. A default install is.

Re:Interesting, but lacking some crucial details..

[Anonymous Coward]

Oi, Smidge you racist, plagiarising, trolling cunt. Fuck off.

Interesting...

[Bert64]

Interesting how much quicker Ubuntu is, especially since the sun java should be built off the same codebase...

I was always under the impression that Intel graphics chipsets were well supported by Linux and had faster opengl performance than windows running on the same chipset, at least some games like quake3 seemed considerably quicker on linux when using intel chips, or is it OSX that has lousy intel video drivers?

It would be interesting to add XP into the mix and see if java performs any better on that, also a few other non java video benchmarks would be interesting so we could see how much of the performance is down to the drivers, perhaps try doing the java benchmarks using an nvidia card too?

Re:Interesting...

[Shados]

especially since the sun java should be built off the same codebase...

They're not. They're using different builds of java for the 2. The benchmark with Ubuntu was using a newer version of the JVM that has some of the more significant performance related modifications in the history of the platform.

Re:Interesting, but lacking some crucial details..

[Bert64]

While there are many factors, comparing the default settings of two systems is perfectly valid, since thats what most people will be using... Only a few people will spend significant efforts tuning the OS, and linux users doing that are more likely to be running gentoo not ubuntu.

Re:Interesting, but lacking some crucial details..

[godefroi]

That's gotta be the stupidest comment I've ever read. Congratulations.

Programming languages are not IO bound. Operating systems aren't IO bound. TASKS are IO bound. ALGORITHMS are IO bound.

Re:Interesting, but lacking some crucial details..

[bckrispi]

The compiler output *is* the same regardless of platform. Bytecode is Bytecode.

Re: Responsiveness

[colinnwn]

When was the last time you or your company raised a stink to Microsoft about performance or compatibility of something and you got it resolved (and rolled into a Windows Update that went out to all users)?

I bet only Fortune 100 companies can get a call back from a Microsoft Engineer with a resolution to a problem. And I bet there are only a handful of cases where the resolution flowed upstream. This kind of stuff happens every day with FOSS.

And those 99.9% of FOSS users that don't have the expertise to fix something can pay the 0.1% who can, and will roll it upstream. It is exactly because of paid development and industry support resolutions that gets rolled upstream that has driven FOSS to be so successful.

OMG!! /. has an interesting article...

[cjjjer]

Oh wait.....

...nevermind.

Re:Interesting, but lacking some crucial details..

[LWATCDR]

True except that most java runs in a server environment and not even on a desktop. A better test would have been to streamline the system to what a knowlegeable user would do.
And of course the big thing is that they didn't use the same version of Java for both systems. .10 has been a big improvement.

Did you detect /and/ set your sensors?

[Gazzonyx]

Did you run sensors-detect, load the kernel modules, and then sensors -s? I always forget the sensors -s call which sets the sensors that sensors-detect found...

Re: Responsiveness

[abigsmurf]

Probably about as much chance of me complaining to Microsoft about bad performance and getting it fixed as going on a programs dev forums and making the same complaint.

I would get a whole lot less flaming and abuse from microsoft too.

Re:Interesting, but lacking some crucial details..

[maxume]

No, the other compiler.

Re:fast java

[ClosedSource]

Well even more common than the myth of "the year of Linux on the desktop" is the myth of "the current version of Java is no longer slow".

Re:Interesting, but lacking some crucial details..

[ozphx]

Yeah its pretty much assclowing IMO. I just booted my Vista laptop this morning. Process explorer shows 9300B idle cycles vs around 200B for other shit (mostly Opera, Steam, and sony's shitty phone sync). Taking them out we're left with bugger all - around 10B cycles "wasted" by default Vista services etc (well, actually SetPoint, AVG, Daemon tools, but regardless). Most of these aren't actually doing anything, just sitting there having started and are now idle. Looks like close as makes no odds 100% CPU avaliable for their JVM.

So according to this, the OS has fuck all overhead in terms of services. So either the kernel is managing the 50% overhead we saw in the SciMark test - or their tests / JVM are crap.

I'd tend towards thinking the latter.

Re:Interesting, but lacking some crucial details..

[ozphx]

See my previous comment. My Vista install still has everything turned on - and realistically it makes zero difference.

I'd suggest either the JVM sucks, or their tests sucked.

What the world needs

[uassholes]

is a website that summarizes Moronix articles in one page without the ads.

If the ridiculous graphics were converted to text, the whole article would be three paragraphs.

Why only two OSes? Just another multipage, ad filled, low-information slashspam?

Re:Interesting, but lacking some crucial details..

[kdemetter]

True , but that bytecode can't be executed directly.The JVM interprets the bytecode , and turns it into OS assembly code , to run it.

So , while the bytecode is probably exactly the same,what it is converted to can be different.

Also , in the java code itself , different code can be executed depending on the OS . Creating a new File may look the same in Java , but the implementation will be different, depending on the filesystem.

Re:Interesting, but lacking some crucial details..

[king-hobo]

That's gotta be the stupidest comment I've ever read. Congratulations.

and thats saying something

Cheaters !!! Server versus client jvm

[abies]

In addition to using different release of jdk (which is probably a minor difference), I have a strong suspicion that they have used default configuration - which is to run client jvm on windows and server jvm on linux.

Each of those configurations have it's benefits - client jvm is quite good for interactive applications, but for anything number-crunching, like the mentioned benchmarks are, server jvm is the must (think about difference between gcc with no opt and gcc -O3 and add some more on top).

Quick check on my vista box with jdk 1.6.0_11, gives 101 for monte carlo in client mode and 220 for monte carlo in server mode. It looks quite similar to their reports (with jvm version change/different cpu than mine maybe being responsible for rest of difference)

If it is really true, that I'm shocked to see to what length some people will go to prove their favorite OS is better.

Next, I'm expecting browser speed comparison with firefox on broadband pitted against IE 8 on 9.6k modem. After that, 3d performance of ubuntu on latest nvidia versus integrated graphics on Windows 3.11.

Secondary Vista 64 boot : Drivers problems

[DrYak]

When was the last time you tried running Windows 64 bit? {...} Two years later, those issues are no longer a problem.

Well, I still keep a Vista 64 partition to multiboot for gaming.
My experience mirrors that of Bert64 :

My hardware didn't came with vista pre-installed (in fact it was home built - around the time Athlon 64 just started appearing on the market).

By chance, the sound card (Audigy), the Graphic Card (Radeon 3850) and the network controller (Realtek 8169S) are all three correctly recognized and handled. So for the purpose for which I kept a bootable Windows partition - gaming - the bare strict minimum is working. Note that for one piece of hardware (Audigy) I had to use community-hacked drivers (stock Creative Audigy drivers are known for several bugs), and that another piece of hardware is made by the creator of the current 64bit ISA, so they're likely to put much more efforts behind drivers.

But for absolutely everything else (PDA, half of the disk controllers, on-board FireWire controller, PCI TV-Capture card, USB Scanner, USB webcam ...) I'm left without any hope. And that is all hardware that is only a couple of years old (it's not some antique piece of hardware whose company went belly up a long time ago - I could understand for older things like no vista 64 driver for 3DFx card or dropped support for 5 1/4" floppy and analog game port).

By chance I mostly run Linux and all this hardware is very well supported out of the box on my distribution.

Or was your comment sufficiently advanced satire that I'm missing? :)

Well it was a half-joking introduction to my more serious rant about lack of Linux 64bits benchmarking for server applications.

Re:Interesting, but lacking some crucial details..

[Anonymous Coward]

Programming languages are not IO bound. Operating systems aren't IO bound. TASKS are IO bound. ALGORITHMS are IO bound.

Please explain how programming languages don't contain algorithms and Operating system's don't perform tasks.

Re:Interesting, but lacking some crucial details..

[shaitand]

There is plenty of Java on the desktop.

For instance these results give insight into why programs like Vuze and OpenOffice.org run so slowly on windows relative to *nix.

Re:Interesting, but lacking some crucial details..

[techprophet]

It is very stupid to attempt to install Windows after Linux. Believe me, I know from personal experience. It takes many hours longer to install Windows 2nd. This means that they probably had Windows as Partition 2 (1 being the recovery partition), and Ubuntu on Partition 3.

Re:Interesting, but lacking some crucial details..

[techprophet]

That is part of the point. Java on both platforms is very similar (though not exactly the same), so the test is more of 'Which OS is faster?' than 'Which Java is faster?'

Re:Interesting, but lacking some crucial details..

[techprophet]

There is a reason both of you posted anonymously, you know.

Re:Interesting, but lacking some crucial details..

[Anonymous Coward]

that maybe so
but it's nice to trim down services (it did make a difference on xp )

because then you're on the path of following
the configuration of least priveledge.
eg your machine is configured to do just what you do
and you do know that like 40% of intrusion is done by misconfiguration by trimming it down there is less chance for misconfiguration to occur ...

Re:Interesting, but lacking some crucial details..

[LWATCDR]

OpenOffice mainly uses java for the database functions, a few file export functions and the media player functions on Unix like OS's. It also uses it for some wizards.
It will not cause OpenOffice to run slow on Windows.
And Vuze is a bit-torrent client. That will tend to be IO bottlenecked. Again Java will have next to nothing to do with it's speed.
But that aside notice I said "mainly" used on the servers. Mainly != only.
Hack I develop in Java. It is a very nice RAD system that runs on Linux, Windows, and OS/X. I would love to see it used more but again this is a benchmark and you should try to make a reasonable effort to make it fair. If not it is no better than saying Windows SUX dude!

Too bad...

[Anonymous Coward]

...that Java is near meaningless and useless nowadays...

Check the facts ...

[xerces8]

I posted this on the moronix forum:

Phoronix discovered that the server HotSpot VM is about 60% faster* than the client VM.
Guess what, this is true on both Windows and Linux.

The real result is not "Linux is 60% faster than Windows" but "server VM is 60% faster than client VM".

Anyone who has doubts, can repeat the tests on their own PC.

* - depends on actual code; for Bork File Encryptor it is about 60%

And this on another forum (no need to translate to see the essence):

Evo, gehajmnis geluftet:

Java 1.6.0_11-b03
HostSpot 11.0-b16

Test : Bork File Encrypter v1.4
Ukaz : java -server/-client "-Dbork.password=testpassword" -cp bork.jar org.matthew.bork.Bork ubuntu-8.10-desktop-i386.iso
Testni fajl na FAT32 particiji. V vsakem testu na isti.

CPU: Intel Core 2 Q6600
RAM: 8GB DDR2 800 5-5-5-5

Windows XP SP3 :
- client VM : 15,5 sekund
- server VM : 8,9 sekund

RIPLinux 7.3 (jedro 2.6.27.7)
- client VM : 15,8 sekund
- server VM : 10,3 sekund

Torej so pametnjakoviÄi na moronix-u ugotovili, da je server VM 60% hitrejÅi od client VM.
Pol so to zlagali v "Linux je 60% hitrejÅi".

Re:Interesting, but lacking some crucial details..

[ultranova]

True , but that bytecode can't be executed directly.The JVM interprets the bytecode , and turns it into OS assembly code , to run it.

Into what ? Assembly is processor-specific, the OS only affects which system calls are available and how they are made.

So , while the bytecode is probably exactly the same,what it is converted to can be different.

There will be small differences in how files are read, for example, but they aren't significant. The real differences lay in the OS side of the calls.

Re:Interesting, but lacking some crucial details..

[shaitand]

While I agree that a tuned install would be an interesting benchmark I still do not agree that testing the systems as shipped and intended to run by the respective manufacturers is 'unfair' simply because one manufacturer has a poor idea of how you should run it.

I'm not going to debate the technical specifics of the apps I used as examples. The point being that there is more java on the server but plenty on the desktop as well.

Even if it were only the server we are looking at default installs are STILL a reasonably fair benchmark. Admins are SUPPOSED to be competent users who know how to tune the system, in the real world competent admins are rare.

'If not it is no better than saying Windows SUX dude!'

At what point DOES it become fair game to say 'Windows SUX dude!' How many performance, security, greed, incompetence, and UI issues is the threshold? Must be fairly high.

Re:Interesting, but lacking some crucial details..

[plague3106]

Well, similar isn't good enough for this kind of test. Especially given that two diffrent versions of java were used.

And the end result seems questionable in my mind; do I care if DOS is faster than Windows? Windows is slower because it has security checks in place. Windows has more fine grained control over security than Linux (ACLs to control access, vs. the normal user / group / world). So ya, one may be faster, but it may be faster over a feature I care more about than performance. Did they even attempt to include that in the analysis?

Seems like it's just a quick cheap way to bash Windows, more than an experiement to show something useful.

Re:Interesting, but lacking some crucial details..

[techprophet]

Windows has more fine grained control over security than Linux WTF? I can make a completely unprivileged user or 2nd root (almost) in Linux. I can fine tune the control scheme to prevent anyone but me and a couple of other users from accessing specific files and directories. There is iptables. And I can modify the kernel source if I want to get deeper control. Can you do that in Windows? (If you can I want to know how!!)

Re:Interesting, but lacking some crucial details..

[plague3106]

Can you make a second full blown root user? No, not out of the box. You can also make unprivledged users in Windows as well, so I don't see your point there. Yes, I know all about user and group settings in Linux. Without creating a whole new group, can you give yourself a group of others EXCEPT bob access to a directory? I've not found a way to do that without ACLs. I know linux can support them... but it's not baked in by default. I never found an easy way to get support, and even if you can do it for files, that leaves a lot of other appliction areas that don't know about ACLs at all. (Think access to objects im memory).

I never mentioned network abilities at all, so I fail to see the point of mentioning iptables.

Yes, I certainly can modify the source. And then maintain my changes, test them with every kernel release, and have all the burden of being a developer of my own special OS fork. Or I can use Windows, which has the features I want already built in, and still provides hooks for things I need to plug in.