Slashdot Mirror
Sun Solaris Vs Linux: The x86 Smack-down
JigSaw writes "Tony Bourke put together a long article, benchmarking File System, System, Compilation, OpenSSL and Web Performance for both Linux and Solaris on x86 hardware. While SPARC's Solaris is said to be more optimized than its x86 counterpart on the other hand so is Linux 2.6 compared to 2.4. Solaris-x86 performed well in the tests, but Linux 2.4 seems to win most of the tests and the overall impressions."
I see nothing in the article about a steel cage. You call that a smack-down?
Was Sun really serious about Solaris on x86?
The fact that they simply "give away" the OS for cheap (i actually got my copy for free from Sun) kinda makes me think they've only released the x86 version to make it "available" to more poeple.
The more people that are familiar with Solaris, in theory, the more Admins/IT staff will end up recommending SUN hardware/software at their workplace. It's a marketing strategy. Not a pervasive strategy, but a strategy nonetheless.
If you take my meaning, mr. Frodo.
do() || do_not();
Guess that question will be asked, and to those to lazy to RTFA, here's his reply :
I chose RedHat 9 for the simple fact that it is a very popular distribution, and is ubiquitous in terms of corporate and personal deployment. Of course it is not the end-all be-all of Linux distributions, but it's both popular and effective, which makes it appropriate as an evaluation platBesides, most of what I evaluate has more to do with Linux itself, and not the distribution. The only significant affect RedHat has on this evaluation is the specific version of the kernel (2.4.20-20.9) and the use of RPMs (which some other Linux distributions use as well).form.
...from the distant past, there's this Slashdot thread from way back in 1999.
There's a "Summary of Points" post a ways down that page that nicely encapsulates most of the discussion.
The Army reading list
In my experience the majority of Developers don't have an in depth knowledge of OS and Hardware performance. The System Administrators almost always have a much better understanding of OS performance. The majority of developers I have worked with are Java developers. Perhaps it is different for other languages. However, shouldn't this be under a Systems Admin category?
This is a surprisingly good article for OSNews. Usually their reviews are limited to utterly trivial things like what the reviewer thought of the default colour scheme, or how easy it was to change the desktop wallpaper. But this one actually has some useful quantitative data in, and refers to things that workstation users actually care about, such as compile times. Whoever this chap is he should take over doing all the reviews from the girl (can't remember her name offhand) who usually does them, because she is pretty much clueless.
The first thing I do when I get a Solaris system is to install a whole heap of GNU utilities, all of which come with any of the Linux distribution.
Solaris/Intel is just a toy that grabs a few extra customers that Sun would have lost otherwise. Boy, you should see it when linux noobs get their hands on it. They get really angry when you tell them "your hardware must be listed on the Hardware Compatibility List". I've seen venomous diatribes directed at "sucky" Sun and its "sucky" OS for not having video drivers for whatever the most expensive game-playing graphics board is these days. And if they actually get the system to install and they see CDE...oh man.
I don't hang on #solaris any more, but damn we would get the same reactions over, and over, and over about Solaris/Intel.
Shutting down free speech with violence isn't fighting fascism. It IS fascism!
I don't think this is Sun bashing.
It is widely known that Sun has waffled on x86. Now because their market share is being eaten up, they are reconsidering years of mistakes...I was one of those admins that had to struggle with Solaris x86.... I woke up though, and went with GNU/Linux + *BSD.
Can Solaris compete with Linux/*bsd on x86? Try a modern distro, and form your own opinions...
Many papers make this mistake. If you ever see scalability comparisons without pure time comparisons, don't trust the results.
Modern != common usage
I'd wager there are many more of these types of systems running important tasks than there are bleeding edge ones. The dot-bomb made people look at the bottom line ya know.
Trolling is a art,
Solaris's strength lies in scalabilitly.
Very true, it's because damn near everything is threaded. Threading is highly encouraged by Sun when programming for Solaris on SPARC.
Each process has threads, if it's a single threaded application it counts as one thread. Each thread is attached to a LWP or Light Weight Processor. The kernel then schedules the LWP time to run on the real CPU.
What's the end result of this? Solaris scales very well on boxes with tons of CPUs because everything is threaded, and some processes have tons.
actually freebsd does run on multiple processors. You can find more info at http://people.freebsd.org/~fsmp/SMP/SMP.html
Evolution or ID?
> You're comparing apples and oranges,
Huh? Mac wasnt even in the article!
Talk about RTFA...
Who bothers to use Solaris on x86? Oh yeah, idiots who don't understand the 'right tool for the job' philosophy.
But what if Solaris x86 is the right tool for the job?
It may not be often but it is at times.
Neither Sun nor Sun resellers really push Solaris x86. This includes during their presentations to customers interested in high performance x86 clusters solutions. Yes, the alternative is always there, showing the Solaris logo along with a RedHat one under available operating systems. Even with clients where Solaris x86 might make sense, Sun salespeople skirt around the issue of O/S and never press their own version. Aside from Sun support (which IMO is really good), would there be any benefit to switching to Solaris? Everyone knows that it's not a core product or moneymaker for Sun, even their own sales associates who definitely know which products to push, and which to let slide. For now, I would leave Solaris x86 as a novelty, at least until Sun itself proves it has feature enhancements outperforming a Linux installation (especially on their own hardware).
"The first thing I do when I get a Solaris system is to install a whole heap of GNU utilities, all of which come with any of the Linux distribution."
They come with the Solaris distribution as well. Not Sun's fault if you don't install them.
A.
...bringing you cynical quips since 1998
it's got the ancient vi
Solaris works for consistency and having a plane jane vi might be a good thing even if vim is better.
The awk/nawk are ancient
I don't use awk often, what's up with them?
and it doesn't ship with perl (last time I checked). Ditto for most of the Unix shell programs.
Yeah it ships with perl since Solaris 8, same with bash, tcsh and zsh.
The first thing I do when I get a Solaris system is to install a whole heap of GNU utilities, all of which come with any of the Linux distribution.
Yep, they're nice to have and don't take up much space if you pick and choose correctly. Here is everything you'll need as far as OSS utilities on SPARC/Solaris are concerned. They can be downloaded as a CD image or individually and are in Sun's package format.
Processor (2) Intel Pentium IIIs at 600 MHz, 256 KB cache
Motherboard Intel L440GX+
RAM 512 MB PC133 ECC
DISK (1) 9 GB Maxtor SCSI
I bet Solaris is designed to run on more serious hardware. I bet DOS apps will run even faster than Linux on this box, even without taking advantage of dual processor.
= 9J =
Meanwhile Solaris 9 for x86 (aka SunOS 5.9, as the article says - it misses the point that Solaris simply means (in Sun nomenclature) SunOS plus the windowing environment, and it once means SunOS plus openwindows, and that Solaris 1.x is SunOS 4 (BSD-based, mentioned) and Solaris 2.x is SunOS 5 (SVR4--based, which is not mentioned directly that I recall)... Er, where the hell was I? Solaris 9 is not available in a version well-optimized for x86. Because you can only relink the kernel and not recompile it, since source code is not provided, it is doomed to this fate. Redhat 9 is also something of a standard, and it happens to come with (and only support) linux kernel 2.4.20. 2.6 has many optimizations but is new. So he mentioned it because it puts both distributions on somewhat equal footing, and in fact in most benchmarks (which are overly simple, but anyway) the systems came up with similar performance when realistic options were utilized.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
Stepping outside of the Solaris --vs-- Linux article, I just didn't come away very satisified. It looks like the author only performed a few shoestring tests while I was expecting an exhausting barrage of tests. IMHO I wouldn't take anything to heart from reading this article. Maybe if the author goes back and expands the number of tests it would be an interestig article. Just look at all the comments from our fellow posters. There are so many people pointing out various issues that the tester neglected. I think the author should take all the feedback and perform a new test with an imporved set of criteria and hardware platforms.
Solaris for x86 has always been for training or utility. Cheap or free x86 versions mean people who want to eventually admin Solaris on Sparc equipment can easily get ahold of it to try without having the capital to purchase a Sparc just for learning purposes.
The other common job for it has been as a utility/admin machine. When you have 20 Solaris Sparc servers to admin, it's just easier and more consistant to also run Solaris on the admin's workstation. Why throw on another *nix with it's own individual setup and quirks when you can run something integrated and consistant. It doesn't really matter if the admin's box is 20% slower than an equivalent Linux machine, it still works fine to admin the big iron.
For some jobs, it is the right tool. Except to idiots who don't understand the job.
at least he was lucky there where drivers. Try installing it on real server hardware, then even FreeBSD has more drivers for stuff like RAID controllers etc. Look at the Hardware Compatibility List.
If you are planning on installing Solaris in your enterprise enviroment you have to buy hardware that will work with Solaris x86 rather than the hardware you normally use. But then again, in a enterprise enviroment, I guess one would choose the Sparc platform or buy the Sun Intel hardware.
I guess he doesn't know that Sun generally releases a T-Patch relatively quickly so that admins can get immediately relief while testing out the real patch.
Ugh, why does everyone assume installing Apache, PHP, and MySQL is so hard on Solaris (or IRIX or AIX for that matter)?
Once you have your GNU environment configured, it's a simple matter of compiling. I haven't run into a snag doing this in over 3 years on three different commerical unices.
Here's a good link for the total newbie:
http://ampubsvc.com/~meljr/AMPS.html
I suppose you could also go to sunfreeware.com (or for IRIX, freeware.sgi.com), but learn to build the stuff yourself and you'll know what's going on, have the latest versions, and have way more flexibility. Isn't this why you're using u*nix anyway? For the flexibility? Don't let the lack of a precompiled ready-to-install package get in your way, you're not stuck in the Windows world anymore.
(end rant)
Solaris does have a few areas where they have done a fantastic job.
For example, when it comes to debugging threaded applications, and having a reliable debugger, they beat us every single time. This is a mix of debugger support, kernel support, libraries support and god knows what else.
Their thread implementation is also very robust. I have no clue about their performance, but I know that you can depend on their implementations being robust. On Linux plenty of thread-related issues are still flaky (big progress being made there), but today, I really wish I had Solaris to debug a few problems.
And there are tons of other little things they get right. My suggestion is that we should focus on what is wrong in our platform, and focus on what is good in their platform, to find out what needs to be solved.
Miguel.
Most slashdotters won't understand or agree with this, but the large bulk of Sun's customers appreciate the fact that command-line options do not mutate over time, that the default behavior of the -foo switch is now reversed, etc, etc. The GNU coreutils maintainer has been busily ripping out all kinds of traditional functionality in the name of POSIX standardization, which would normally be a good thing if he hadn't gone way too far. (I don't give a fuck if "uniq" and "head -1" aren't full POSIX, they're in my scripts, they're in my head, and they're staying there.)
If Sun tried to make as many incompatible changes to their core utilities as the GNU utils does, somewhere upwards of 80% of the customers would just walk away.
Yes, I install GNU coreutils and all kinds of happy stuff (like a decent shell) as soon as I open up a Sun box. But I leave their versions in place so that old PATHs still get the behavior they expect to find. Everyone here loves the cutting edge, and loves to cut down anyone using version ($latest-1). Sun's primary customers aren't like that. They want stability in the core utilities across years, not new features every few weeks, or even months.
(Yes, the last 3 or 4 versions of Solaris have all shipped with Perl. It's a slightly older, stable version of Perl. There's a bunch of stuff on the freeware companion CD too, as well as sunfreeware.com. Those who want a stable Solaris get it by default. Those who want bleeding-edge tools can easily download the packages.)
You cannot apply a technological solution to a sociological problem. (Edwards' Law)
"The new TCP/IP stack - code-named Fire Engine - has 10 gigabit and 100 gigabit Ethernet networks in mind." Available for testing download now
Sun is quoted:
Bill Joy doesn't even use vi anymore.
The rest of you should take that as a hint.
At least RMS eats his own dog food (emacs).
A Pirate and a Puritan look the same on a balance sheet.