Not logged in
Log in now
Create an account
Subscribe to LWN
An "enum" for Python 3
An unexpected perf feature
LWN.net Weekly Edition for May 16, 2013
A look at the PyPy 2.0 release
PostgreSQL 9.3 beta: Federated databases and more
SGI Altix 3000 Proves Favorite With Linux Journal Readers
Posted Nov 6, 2003 22:25 UTC (Thu) by mattdm (subscriber, #18)
Posted Nov 6, 2003 23:02 UTC (Thu) by joeman (guest, #6711)
Posted Nov 6, 2003 23:51 UTC (Thu) by vblum (guest, #1151)
Sorry to ask so bluntly; but I'd honestly and sincerely appreciate any opinion ...
Posted Nov 7, 2003 18:20 UTC (Fri) by ogre (guest, #14142)
The Beowulf style clusters you are talking about require a work load that can be broken down into discrete units. For example making a movie each system can be handed a frame or set of frames to render and doesn't need to know what's in the next frame.
In some simulations and mathematical models the calculations cannot be broken into discrete units because there is too much interdependancy. I've never worked with this sort of thing but I think things like automobile accident simulations, simulations of nuclear bombs, or virtual wind tunnels cannot be broken up into discrete work units so a beowolf cluster would do little good.
Perhaps my examples are off whack, like I said I've never worked with these type of systems.
Posted Nov 7, 2003 4:38 UTC (Fri) by mattdm (subscriber, #18)
Posted Nov 13, 2003 6:46 UTC (Thu) by cinnerz (guest, #5943)
They are actually pretty nice machines. A lot easier to manage a single 128 way machine than a whole bunch of little machines.
Posted Nov 7, 2003 3:34 UTC (Fri) by horen (subscriber, #2514)
I don't know about going's-on in other countries, but here in Israel, SGI has repeatedly engaged in a version of "dumping" -- selling brand-new equipment as "refurbished" -- in order to come in as low-bid on numerous tenders within the university community and sell less-than-cutting-edge equipment.
In so doing, the High-Performance Computing Unit of the Israel Inter-University Computation Center, as well as individual universities, has been "stuck" with:
SGI's dubious business practices duped management into purchasing the new Origin 2000 (item #2), rather than investing thse same funds to upgrade both the existing Origin 2000 (item #1) and the heavily-used Cray.
And SGI has done it, again, "dumping" a brand-new TP4100 storage unit as "refurbished" in a tender for a centralized bioinformatics unit, in order to breathe new life (hah!) into the Origin 2000 (item #2); this, despite far better solutions from Network Appliance and other vendors.
While I realize that P.T. Barnum's famous "Sucker born every minute" applies no-less to academia, it is a sad commentary when one of Silicon Valley's (once) flagship companies chooses to conduct its business in this fashion, rather than on the merits of its products.
One final thought: if SGI had been a more "socially responsible" firm, it would have counselled upgrading the Origin 2000 and Cray, rather than what amounted to "throwing good money after bad". Shame on them, for turning a win-win situation into a circus sideshow, where the biggest losers are the students, researchers, and the Israeli taxpayers.
PS: SGI recently sold an Altix 3000 to the Weizmann Institute for Science -- we'll see if their researchers stop using the Cray and/or local clusters.
Posted Nov 7, 2003 4:41 UTC (Fri) by vblum (guest, #1151)
Customers know this; their sales people know this. IMO, SGI's sales people must be completely unable to sell anything to anyone for the asking price. So, they resort to the only thing they can do: Offer useless equipment (which is not anything they'd really want to sell in the first place), but at conditions which are at least tolerable for the customer.
I have heard an Altix price quote above $10.000 per CPU, Itanium CPU (and that's a lie - it was very well above). SGI has fabulous technology and engineers ... but at that price tag, even the very best scaling is beaten hands-down by a rusty old x86 cluster - then, at least, you can run single-cpu on multiple machines ...
Trapped like that, it's no wonder that SGI's sales staff is desparate enough to resort to somewhat shady deals ...
Posted Nov 7, 2003 16:36 UTC (Fri) by parimi (subscriber, #5773)
The Origin 2000 we had was a great machine. Perhaps only corporations with lots of money can help SGI survive in the market..
Posted Nov 7, 2003 9:57 UTC (Fri) by evgeny (guest, #774)
The Cray has been idle for a couple of years. Not that it was _efficiently_ used ever - one has to use specially tailored libraries and/or write code suitable for parallelization to utilize the Cray's power. Having spent a nearly-six-zeros bunch of bucks on the hardware, some essential HPC software wasn't purchased. And in scalar performance benchmarks, my Pentium-100 box w/ Linux at that time (5 years or so ago, when it was bought) beated it slightly ;-). Of course, it (Cray) had a couple of GB of RAM which wasn't so easy to get around on a PC back then.
In general, decisions about purchasing new hardware (and more generally, about anything computing-related) here at the Weizmann are often taken in a way that I'd believe could only be possible under communist regimes. To have an opinion in the matters and expressing it openly results in closed accounts, firing off, etc. This concerns both researchers and computing center staff.
Posted Nov 7, 2003 10:00 UTC (Fri) by evgeny (guest, #774)
Should be read "for vectorization and parallelization".
Copyright © 2003, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds