Slashdot Mirror
SGI to Build Commercial Linux Supercomputers
jfinke sent in a link - as did many others - to a brief Yahoo News story about SGI's plans to move into the parallel processing supercomputer marketplace with Linux as their base OS. "We're really pumped up about it," said Beau Vrolyk, senior vice president of SGI's
product group. "It represents the beginning of a whole new generation of
supercomputer."
So, the article says that SGI did a demonstration with Itanium chips. Is this true ? I thought that Itanium was only available in simulators ( that's how they're developing Linux/IA-64 ), no silicone yet.
Anybody care to elaborate ?
Sorry guys, but I believe SGI needs some mindshare, and that's why they are doing this. Irix scales much better, and so does Solaris and UnixWare. But, granted, they are not as "hot" these days.
It's a good thing for Linux and SGI, but not because Linux is such a good SMP OS.
OT: I have just read they are selling Playstation here in Helsinki for 700 FIM (~145 US$) Do you think it's a fair price?
Sigged!
Actually, though the comment doesn't add much, it's actually highly relelvant. However, what the heck would you call a cluster of these things as they're already Beowulfs? Megawulf?:)
Bill - aka taniwha
--
Leave others their otherness. -- Aratak
I think IBM's still making money selling the descendants of that line of computers they came out with in 1963 or so (System/360's descendant, the System/390), as well as the descendants of that line of computers they came out with in the middle or late '70's (System/3x's descendant, the AS/400), so I don't expect that to happen in the near future.
If customers are willing to pay them money for boxes that aren't Linux machines running Apache, it's not clear that it'd make sense for IBM to refuse to sell them those machines. Perhaps those are all legacy machines, but it may still cost somebody less to buy a non-Linux (or non-UNIX in general) machine than to convert their application right now. (Besides, I wouldn't be surprised to see Linux and Apache - along with Windows and Solaris and HP-UX and Digital^H^H^H^H^H^H^HTru64 UNIX and... and IIS and Netscape Enterprise Server and... - become "legacy systems" some day....)
I am a little wary that this may just be a publicity thing. SGI has spun it's Cray division, and wasn't that in part because it wasn't as successful at they had hoped.
Linux may be full of hype right now, but if they weren't successfully with an already established supercomputing platform (ie. the Cray), how will they manage with something new.
On the bright side, even if their endevour fails, perhaps we'll get some cool new (and hopefully GPLed) stuff ported over to linux.
Dana
A lot of SGI's products (the open source ones) are aimed as toolkits for system management, or are aimed at increasing performance and stability (XFS).
One such project is a toolkit for obtaining system performance statistics. They write the toolkit, and give it away for free, and then they write the management tool that sits on top of it, and sell that!
And how better to know just what your cluster of machines is doing, than with monitoring software? And produced by the same company that sold you the hardware? Why not?
By giving away the toolkit (and it's source), you end up with free improvements, free ports to other platforms, and toolkits that monitor stuff you may never have even thought to monitor in the first place, broadening the scope for your product.
Even without SGI, Linux is becoming more scalable with each release. Now SGI is investing real money in the Linux kernel. They're working on projects aimed at implementing the best features of Irix under Linux. It won't happen overnight, but it won't be that long either.
It makes a whole lot of sense financially (especially for a company that needs to cut costs). They can let the Linux community take care of all the mainstream features that every OS must have, and SGI can focus on the handful of things that are really important to their particular market. Even if SGI ends up being the only maintainer of ccNUMA for Linux, it's still cheaper than having to maintain the entire OS.
The other exciting thing about Linux is the mainstream applications that are emerging. I know people don't buy an SGI box to run spreadsheets, but it's nice to know that you don't have to keep a Windows PC around just for the odd time you want to run an office productivity application. Linux lets SGI ride the wave of new mainstream developments while still catering to their niche market.
It is tempting, if the only tool you have is a hammer, to treat everything as if it were a nail. - Abraham Maslow
Irix is a much more robust and scalable system than Linux. This may change with time, but for the present, it is true; Linux is still lacking in the multi-processor area. And as for Enterprise, well - Irix is proven in the enterprise, whereas Linux is still be looked at as less-than-ready.
-----------
"You can't shake the Devil's hand and say you're only kidding."
I think this is an important point.
When NT started to (self-declaredly) encroach on what had been UNIX workstation territory 3 or 4 years ago, one of the arguments that MS made on its behalf was that people wouldn't need to have a separate box on their desk for common office (read "MS Office") tasks.
This aspect is one of the things that was hyped in The Analysts' reports (Brown? IDC?) finding lower TCO for NT than for UNIX machines. Now that there are some Web-based application suites semi-available (and more poised to be really available RSN
StarOffice, for Free? Heh! Running on a multi-processor SGI sytem? That sounds just about right
timothy
p.s.(And one day there will be a funny graph of IDC opinions, with an inflection point around this year
jrnl: http://tinyurl.com/c2l8yr / foes: http://tinyurl.com/ckjno5
I've got a TNT card in my PIII, running Linux, and while it's a pretty nice setup, it's not a replacement for an SGI system. Just having a Mesa setup won't do it. You need the software and Linux just hasn't got it yet.
As for the benchmarks: Well, I trust Linux benchmarks by Alan Cox about as much as I trust NT benchmarks by Microsoft.
-----------
"You can't shake the Devil's hand and say you're only kidding."
> We should see more strength in the PC SMP for linux, *bsd areas.
The article was pretty lean on technical details, but it did say it was talking about a cluster. I would guess they're going the Beowulf route.
--
It's October 6th. Where's W2K? Over the horizon again, eh?
Sheesh, evil *and* a jerk. -- Jade
You write: Irix scales much better, and so does Solaris and UnixWare.
... devastating to anyone that doesn't jump on the bandwagon that is. Development and post-sales maintenance/support are extremely costly overheads for any manufacturer, so a decision which massively reduces that expense makes great economic sense.
Whatever systems happen to scale better today, Linux and the free BSDs will scale better than anything else tomorrow, simply because there are 10 zillion developers beavering away at them continuously.
Furthermore, this continual and rapid improvement will happen regardless of whether or not SGI invest any development time of their own. The economics of this are devastatingly obvious
Furthermore, with Irix behind them, SGI have the pedigree to stand out in the Linux marketplace despite the eventual dilution created by GPL licensing of their enhancements. I doubt if they'll lose many of their old customers, and they are bound to acquire many new ones from their new and greatly expanded Linux audience.
With this move, I reckon that SGI have their future assured at least until the competition wises up and catches up, and as long as their marketing, pricing and distribution is similarly forward thinking. But that of course remains to be seen.
"The question of whether machines can think is no more interesting than [] whether submarines can swim" - Dijkstra
I cannot say with certainty that this article is about the same machine I'm thinking of, but...
SGI recently finished a Beowulf for the Ohio Supercomputer Center, and has put it on display at SC99 before actually delivering it to OSC. 128 Xeons for computation (32 SGI 1400L's, each with 4 Xeons, with one more 1400L as an admin node; Myrinet interconnect).
Since Itanium hasn't actually been released yet, I expect anything anyone debuts in the near future built around Itanium is only debuting 4-color glossies. But, that is the first step before debuting an actual product built around a brand-new processor.
Christopher A. Bohn
cb
Oooh! What does this button do!?
-----------
"You can't shake the Devil's hand and say you're only kidding."
No offense intended here; this comment is meant to be about 25 deg C, and is just a clarification for what seems to be a common misconception.
No application can make use of a parallel computer without a programmer first identifying the inherent concurrency. There are pre-processors that attempt to do this for you, such as BERT 77, but IMO that cannot replace the intuition and basic understanding of the problem that a human programmer provides (with all due respect to Doug & everyone else at Paralogic). Once that concurrency is identified, then implementing it using the MPI or PVM API, or using HPF directives, or using pthreads (depending on your platform) should be fairly straight-forward.
Computers are not magic! You cannot feed a program written for a sequential computer into a parallel computer and expect speedup, period. At run-time, the computer just doesn't have sufficient view of the application to even attempt to identify the concurrency for you, and it certainly doesn't have the cycles to spare to decide what aspects of the concurrency should or shouldn't be parallelized. The best you can do if you won't or can't parallelize your code is to submit multiple instances of the application, with different inputs, to the batch scheduler and obtain a greater throughput.
Christopher A. Bohn
cb
Oooh! What does this button do!?
Last night the Vancouver Linux users group presented a talk by Dr. John Mashey from SGI. It was a very interesting talk, and convinced me that SGI does "get it" and is truely dedicated to Linux and the open-source concept.
I don't think that MIPS/Linux is a priority for SGI. They see MIPS/Irix at the 'top' of the market, and Intel(x86 and itanic)/Linux at the 'bottom' but moving up over the next few years. They know it's coming, and they'd rather spend resources to influence its development than be a passive by-stander as they are with NT. So, they feed features and ideas from IRIX into Linux, while continuing to sell and maintain the high-end IRIX systems as long as there's still a market (which consists more of corporate uses like automotive crash-test simulation than of TV and movies).
He also said that several of SGI's customers want Linux, and want certain features in it (like raw I/O for big databases). So, SGI either helps develop this and keeps the customers, or loses them.
The XFS filesystem is one of their major contributions, and he said that one of the main delays there is that they are going through the code to remove any sections that could cause patent headaches in the future. One hopes the result will be slightly more usable than the initial Mozilla...
---
disclaimer: these were my impressions of Dr. Mashey's talk; they may or may not actually represent his views, which may or may not be those of SGI.
as another nail in SGI's coffin? This will get them some attention but unfortunately not get them much business. I don't want Linux on a super computer, it would cease to truely be linux if you changed it enough to actually work as a super computer. What makes a computer super is not it's total operations in a second but how efficiently it handles those operations. Crays use(d) vector systems, the chips are mostly SIMD circuitry which means you can get more usable data from a single instruction. Until recently PC class chips had never touched SIMD, they were all SISD circuitry. This means the chips in Crays were the opposite of what a PIII is today, it was mostly SIMD circuitry with small SISD units (where the PIII is mostly SISD with one or two SIMD units). Besides the actual hardware difference, Linux just does not have the super computer mentality. Irix was designed to scale on hundreds of processors and address hundreds of gigs of RAM, Linux makes poor use (comparitively) of only hundreds of megs of RAM. If Linux were tweaked by SGI's old Irix people (do they even work there anymore?) it would really cease to be Linux. You have to remember that Linux was originally designed to be a flavour of unix that ran on PC architecture, not power hungry-underground complex requiring mainframes. This is no kind of slur against Linux it is focused on SGI. It's a shame they have been reduced to trying to get under someone else's PR umbrella to make headlines. If someone offered me the choice of an old Origin 2000 or one of their Linux super comps I would go for the Origin based almost entirely on principle.
I'm a loner Dottie, a Rebel.
Two big firsts on IA64 for Linux and I think it's a great indication that Linux will be the first released OS for IA64.
Some more info:
sgi's announcement
Project Trillian which is porting the linux kernel to IA64.
Note that what you are proposing is not all that simple: _increasing modularity_ and reworking the _SMP_ scalability of the kernel invovles _major changes_ to the kernel structure. Here lies a great risk - that the changes will be so sweeping, that the linux kernel structure will fragment: we'll have the "Linus" kernel, and the "SGI" kernel. Hopefully if this happens, the changes will merge into Linus's kernel quickly, otherwise if SGI release their changes to quickly we'll have irreversible fragmentation.