Slashdot Mirror
64-bit Solaris Tests Successful
mulan writes "Following a successful email sent from a Solaris64 box, Sun announced today that the Solaris Operating Environment is running on engineering prototype systems based on Intel's Itanium processor. The press release is on line, while meatier details, white papers and documentation are available at the Solaris64 developer center. "
I don't doubt that, Bruce, but I seriously doubt you got a Xeon with 2MB cache. I'm seeing $3680 as the minimum price on the net.
Sparc hardware is nicer than most Intel hardware...if you remove cost from the equation. But mostly I've found that Solaris users who dump on Intel Solaris are either just blind Sparc bigots, or formulated their opinion on an ancient version of Intel Solaris when it really was a terrible product. The Sparc bigots are really reflecting on Intel hardware, not the Solaris product. When pressed, I've found they will generally admit that there isn't really anything wrong with Intel Solaris. The past-bad-experience people just need to get with the program and realize that Intel Solaris has matured considerably and is every bit as capable as Sparc Solaris if you don't try to go beyond what the underlying hardware can handle. Just don't use IDE disks with it and expect to use a third-party X server, if you need an X server.
So that is the story with Sparc Solaris users. Linux/BSD users might reject it for ideological reasons, cost (it is free only for non-commercial use), baseline system requirements, and the painfully short hardware compatibility list. In my opinion, anything less than 128MB of RAM just doesn't cut it for a Solaris desktop system. FreeBSD is quite usable in the same capacity with half that.
Along many dimensions, FreeBSD does outperform Intel Solaris on the same hardware (I haven't done many comparisons with linux), but not by a huge margin (except metadata intensive filesystem I/O but rumor has it that Solaris 8 will include FreeBSD's softupdates code). FreeBSD generally has a snappier feel too it for interactive use, but I'm hard pressed to explain that through the synthetic and real world benchmarks I've run.
SMP is the one area where Solaris is the clear leader. Both Linux and BSD need same major internal reworking to get where Solaris is now. Intel Solaris also supports Intel's new address extension thingy so you can have truly staggering amounts of real RAM...if you can find the hardware that supports it.
Minor point: HP-UX on PA-RISC 64 has only been available on servers up until like next month, when HP-UX 11 debuts. 10.20 isn't really a 64-bit OS.
:) (BTW--I have always felt that Mr. Packard should have demanded that his name be first :)
But point taken
My journal has hot
Unless I read it wrong the article specifically said that they announced they were planning on sticking with their *own* chips. Instead of the (more logical path of dropping their 64bit chips for) Intel 64bit chips.
It was a bogus argument that Intel's chip were better then current 64bit chips.
-Brent--
Yep, you got that right. Like Fizgig said, it's a Xeon /w 2MB cache. Those chips are MUCH more expensive. Plus, I kinda doubt the $6500 list quoted on IBM's web page is for the "raw" chip. Probably includes a CPU "board" that plugs into the Netfinity much like Compaq's SMP servers do.
I was a little surprised also, because I had to spec out a E450 running Solaris vs an Intel box running WindowsNT. Much to my surprise the Intel/Windows "solution" was more expensive than the SPARC/Solaris option. It was great to see because I don't think the apps that we were going to run on it would have faired too well with the NT server.
And I did try to be "fair" to the Intel/Windows box by spec'ng out "comparable" equipment. I could have spec'd out the 8-way Netfinity which would have really cost more than the SPARC, but that wouldn't be too fair now, would it?
More like one year, given that this press release has Sun announcing Solaris 7 on October 27, 1998, and given that Solaris 7 was the first fully 64-bit OS from Sun (Solaris 2.6 supports 64-bit file offsets, but that's it).
The title of the original article was a bit misleading, as "64-bit Solaris" has been running for about a year on Sun's SPARC V9 systems; however, I suspect at least some nerds consider it news that it also now boots on Raseodymium or Echnetium or Odium or whatever the hell that name was that Intel presumably spent lots of money to get somebody to come up with....
I recently read Neal Stephenson's In the Beginning was the Command Line (good read, I wonder if that link is a legal copy?) and can't help but think about the probability that without support from Windows' home edition (whenever and whatever it may be), the IA-64 won't become a commodity chip and may be just as out of reach for Joe Hacker as the rest of the chip world (with the notable exception of PPC, of course.)
I'd hate to see that happen. I would love to ditch bloat on my CPUs as much as I loved ditching bloat in my OS :-)
This presumes that there will be "Sun IA-64 stuff", rather than just Sun SPARC hardware. That isn't necessarily going to be the case.
What's even worse is the latest print copy of _ent_magazine_ (The Independent Newspaper for Windows NT Computing) has a front page article on Intel's 64bit chip where they quote: "Intel announced its newest chip, Itanium, which was previously known as Merced. Several OEMs, however, are sticking with their current architecture."
They seem to imply that the OEMs are foolish for continuing to use their own 64bit chips when now Intel has a *real* 64bit processor. As if for some reason Intel is a saviour. Yes, we can now replace our crummy 64bit chips with superior Intel ones. Oh, wow!
I predict we'll see this kind of argument more in the future. "Well, you'll want to use Intel's chip because they are better then others that have been out longer. Nonsense. However, this will make other chips cheaper, thank goodness.
Of course, by the time 64bit chips are common-place the other "OEM's" will be using 128bit chips...
-Brent--
And to think HP could be behind, but one of the key developers to IA-64, and forced Intel to add PA-RISC compatability. If that's true.. that's just a bit sad.
"Open Source?" - Press any key to continue
Sun has been runing 64bit for over 3 years, just becauls Intel has finaly gotten on the band wagen (Which by not I think that we all know) is not news worthy. Unless you are surprised thatsun is continuing to make x86 solaris?
Having tried a SPARC based personal workstation a while ago, I can only say that this is good news for Sun. Their systems may deliver nice I/O throughput, but their processors seem to be outclassed by Intel, especially so when taking into account the price tag of SPARC hardware.
...
I think Sun has 2 problems: Linux, which from a free download gives you a much more polished software install than the factory pre-installed Suns have (and which is certainly good enough unless you're doing realy high-end stuff) and of course the ix86 chips. Intel and AMD might not be there for the really high-end market yet, but for the workstation market they are dirt cheap at similar (or in my case even better) performance than Sun could provide for twice (or even more) the price. Sun's success is increasingly mysterious to me
But IA-32 instructions, as the "32" suggests, don't support a 64-bit flat address space, so, to run Solaris 7 on IA-64, Sun has to treat Tterbium or Otassium or whatever the heck it's called as an IA-64 processor; the fact that it runs IA-32 instructions doesn't help (unless you boot Solaris x86 on it, but all that would demonstrate is "yup, Adium's IA-32 mode works").
The address-space limitations may require a 64-bit processor, but the file size limitations don't, as long as you're not mapping the entire file into your address space - plenty of OSes running on 32-bit processors allow files bigger than 4GB.
If by "double precision" you mean "64-bit floating-point numbers", with the exponent+mantissa+sign taking 64 bits total, what current 32-bit processors don't support 64 or more bits of floating-point number in hardware (and/or don't run software that supports them as well)?
Integer arithmetic instructions that handle at most 32 bits at a time, and 32-bit pointers, don't limit file sizes to 32 bits (one can synthesize 64-bit or bigger arithmetic operations from the integer arithmetic instructions - and most C compilers, these days, will do it for you, and compilers for other languages may do the same); they only limit how big a file you can map into your address space all at once.
They also don't limit the width of floating-point numbers.
The key limitation of 32-bit processors is the address space limitation you noted; that one is a bit more of a pain to work around (e.g., by mapping stuff into and out of your address space manually).
64-bit, or 64-bit with some 128-bit instructions. The Emotion Engine is, according to a Microprocessor Report article, a processor that implements the MIPS III instruction set plus some MIPS IV stuff, with some additional vector and MPEG-2 decoding support added. The operand bus for the vector stuff is 128 bits, but that's not enough to make it a full-blown 128-bit processor (any more than the SIMD stuff in MMX/SSE, or AltiVec, or..., make processors that implement those wider everywhere).
At least not any technical problem. :-) There might be a corporate pride problem (look how long it took Scott McNealy to publicly use that "M" word :-)).
Someone taking my name in vain ? Thats certainly
not my quotes
Solaris scales to 64CPUs (partly because of their
kickass memory bus on the ultrasparc). We beat them flat low end but believe me, for a lot of
things on 8+ cpus it has us hammered. On 32+ cpus
Id be willing to bet it wins aginst 2.4 once we
have it finshed
I wouldnt buy a Xeon for most things either. A
quad Xeon costs the same as a rack of 2U celeron
boxes and ethernet switch. 20 celerons versus 4
xeons, 20 celerons with a total of about 8 times
the memory bandwidth of the xeon box
A couple of points:
"The invisible and the non-existent look very much alike." -- Delos B. McKown
Dear me. Believing a slashdot poster. Want to buy
a bridge, or some dehydrated water ?
Solaris Ultrasparc SMP is very good.
Thats a real quote from me.
Proof that not only MS creates FUD...
/etc/inetd.conf gets commented out. And not starting daemons you don't need helps too.
1. I think you got the two Oses reversed here, as Solaris scales to 64 processors I believe. Anyway, show me where Alan Cox says this.
2. Say it with me, folks:
THE OS IS AS SECURE AS THE ADMINISTRATOR IS COMPETENT!
NO sane administrator would use anything (except perhaps OpenBSD) out of the box. At the very minimum most of
3. The focus shouldn't be "is it Open Source?", but rather "is it the best program that I can use now to fit my needs?"
Linux is a nice system for the "average" PC. But the BSDs are probably best for Web servers, and Solaris is probably best if you need a huge file server or database machine.
Solaris is definitely the wrong choice for home PCs, which have no SCSI and 1 processor, but it's not intended for that. Linux is.
4. The actual situation is more like this: Once the server machine gets big enough to warrant using Solaris on it, the price of the OS is a small fraction of the price of the entire system. And yes, for those machines (4 processors or more) Solaris may perform better (gasp!) than Linux.
5. I must be part of a different revolution than you. My revolution is against Microsoft. My revolution does not say, "Use GNU/Linux OR DIE!!!" My revolution says, "There is more to life than Microsoft products. There are better ways to compute. It's up to you to use the best product for you. It's just that we think you can do better than Microsoft products and would encourage you to research the alternatives."
Open Source is nice, but you have to figure a lot of people just won't _care_ whether they get source code or not, as long as the program works properly.
Glückwünsche, haben Sie Slashdot ermordet, indem Sie zum korporativen Druck beugten und Subskriptionen einlei
It would be a strange world where Sun offers support for Windows 2000 on it's hardware! Especially since Sun touts it's position as the only major computer company without some deal with Microsoft.
What's more likely is that Sun is going to make damn sure that Solaris runs better on Sun IA-64 stuff than on your commodity Dell or Compaq. Some of this will be due to just plain better hardware from Sun, but I'll bet that Solaris will have a few hooks in it that will disimprove performance when running on a non-Sun box.
Business. Numbers. Money. People. Computer World.
First of all Solaris x86, is horrible when compared to Solaris/Sparc
/. opinion is that Solaris/x86 sucks so hard. Certainly Sparc hardware is nicer than Intel hardware, but Intel hardware is good enough for Linux for most of you.
I'm at a loss to explain why the general
Certainly Solaris/x86's hardware support is much thinner than Linux's, and the distribution is pretty thin compared to Linux, but from what I've seen, Solaris on Intel is a very solid operating system once you have it up and running.
Furthermore, in the PC Week "NOS Shootout" a few months ago (which got everyone here up-in-arms), Solaris/Intel was faster and scaled better than either NT or Linux.
Perhaps if Sun wrote a few more device drivers, and shipped Gnome, KDE, the GNU toolset, XFree, and so on, Linux folks would have a little more respect for the OS....
Business. Numbers. Money. People. Computer World.