Slashdot Mirror

Shit! They had the best parties at Supercomputing.

Pratt & Whitney, one of the big 3 jet engine makers, has been doing that for over a decade. It is there primary means of supercomputing.

They have been at it so long that they had to write their own message passing system (PROWESS) because MPI was not there yet.

I used to work for them as a computational fluid dynamicist, we were the main consumers of this "cluster".

This is not an Internet2 speed record (lsr.internet2.edu), which is measured between single host pair. This demo was done for the SC2004 bandwidth challenge using a large number of hosts.

The SX8 uses a 90nm CPU clocked two times as high as the one used in the SX6 respective the erarth simulator. Taking into account all the additional improvements it's save to say the SX-8 will be twice as fast as the old one - so 65 TFlops seem reachable.

For those interested in how vector processing compares with scalar processing in terms of absolute performance and computational efficience, this paper of Leonid Oliker et.al. is definetely worth reading:

http://www.sc-conference.org/sc2004/schedule/pdfs/ pap247.pdf

If you want a good laugh, attend SC2004 this year.

OK, I'm being cruel. But c'mon, if you have a 160 IQ and a Cray X1 is your bitch, why can't you at least trim that beard?! Is there something about shmem that requires vast amounts of facial hair?

So far, they don't compare at all well with various Linux/UNIX implementations. However, I'm sure they'll manage to get their nose under the tent flap where either they paid their way in, or a PHB thinks that it'll leverage his existing stock of desktop machines. My lab built a couple of turnkey systems for the later scenario, where the given reason was that the typing pool cpus could pitch in some cycles to the cluster after hours. What really happened was that the non-com tasked with managing the systems was trained as an MCSE, and he wanted to stick with what he knew, come hell or high water.

The article says that they used ttcp which is a memory-to-memory bandwidth testing program. Most would consider that unrepresentative of reality. On the other hand, today's supercomputers have a tremendous amount of memory (1.2 TB, 6 TB, 10 TB, 33 TB, etc.) so memory to memory is possible.

Others have suggested that disk speeds cannot sustain that rate. However, supercomputer disk arrays can easily keep up (4 GB/s or 32 Gb/s).

Finally, it is possible to achieve nearly the same result (multiple streams instead of a single stream) transfering real data (23.23 Gb/s).

[Bias alert: I am a member of the team that set a previous Internet2 Land Speed Record, Guinness World Record and won the "Bandwidth Lust: Distributed Particle Physics Analysis Using Ultra-High Speed TCP on The Grid" or "Moore's law move over" award at SC2003.]

Now, before you complain that the technology is not available to "mere mortals," let me point out that we first started experimenting with 1 Gb/s Ethernet at work 5 years ago. Now it is readily available at reasonable prices for consumer desktop machines. (Apple has had it standard in G4 desktops for 4 years.) The problem is not with consumer hardware, it is having access to true broadband (not cable modem or DSL), at least in the USA. Although your LAN may support 1 Gb/s, your download speed is limited to 1-3 Mb/s (cable) or 256 -786 Kb/s (DSL). (Your upload speeds are significantly lower.) Since the link provider has very little incentive to upgrade service, I doubt that will change very quickly.

So, yes it is possible. No you can't have it (yet)!

The article says that they used ttcp which is a memory-to-memory bandwidth testing program. Most would consider that unrepresentative of reality. On the other hand, today's supercomputers have a tremendous amount of memory (1.2 TB, 6 TB, 10 TB, 33 TB, etc.) so memory to memory is possible.

Others have suggested that disk speeds cannot sustain that rate. However, supercomputer disk arrays can easily keep up (4 GB/s or 32 Gb/s).

Finally, it is possible to achieve nearly the same result (multiple streams instead of a single stream) transfering real data (23.23 Gb/s).

[Bias alert: I am a member of the team that set a previous Internet2 Land Speed Record, Guinness World Record and won the "Bandwidth Lust: Distributed Particle Physics Analysis Using Ultra-High Speed TCP on The Grid" or "Moore's law move over" award at SC2003.]

Now, before you complain that the technology is not available to "mere mortals," let me point out that we first started experimenting with 1 Gb/s Ethernet at work 5 years ago. Now it is readily available at reasonable prices for consumer desktop machines. (Apple has had it standard in G4 desktops for 4 years.) The problem is not with consumer hardware, it is having access to true broadband (not cable modem or DSL), at least in the USA. Although your LAN may support 1 Gb/s, your download speed is limited to 1-3 Mb/s (cable) or 256 -786 Kb/s (DSL). (Your upload speeds are significantly lower.) Since the link provider has very little incentive to upgrade service, I doubt that will change very quickly.

So, yes it is possible. No you can't have it (yet)!

While I certainly disagree that you can't build a very high performance computer out a cluster of computers (Linux or otherwise), there is a lot of merit to the fact that clusters just don't scale well for certain classes of applications. Hence the renaissance of the vector supercomputer (ala the Earth Simulator ).

Obviously, this guy is plugging the new Cray X1 architecture, which really is quite promising. For instance, check out this paper by some folks at Oak Ridge National Lab that appeared in Supercomputing 2003.

Of course, since this is Slashdot, I expect that there will be a deluge of posts decrying everything about the new Cray machine because it commits the cardinal sin of NOT USING LINUX. Oh, the horror!

This year's Supercomputing 2003 Bandwidth Challenge netted some cool results in this area including 23+gbps inside the US, 7.56gbps between the US and Japan, and 8.96gbps throughput to a remote network filesystem using GPFS. There are even some pretty graphs. My former co-workers at SDSC were involved in a lot of this work.

In case anybody wants it, the link to the conference is at

http://www.sc-conference.org/sc2003/

Several of the lectures are being broadcast via high bandwidth video if
you are on Internet2.

It should be interesting to see where this system will rank. The 22nd top 500 list will be announced this week at the Supercomputer conference. Further details on the top 500 list can be found at the Top 500 list.

To put money into a conference booth, usually means that a company is making some serious consideration for that market.

So here's the trivia question:

When did Apple Computer start being an exhibitor at the International Conference for High Performance Computing and Communications?

Apple's been aware of the Top500 list for a several years.

Supercomputers used to be 32-bit with the Top500 list accepting 32-bit results. At the time Apple started attended the SC Conference, I asked Jack D. if he would accept the 32-bit benchmark results for the Top500.

At the time Jack said no -- adding back 32-bit would be confusing to the community.

It took some effort and many voices to convince Apple to pay attention to the scientific market. It's nice to see them headed towards the Top500.

--Hopscotch

HINT: 9991

The new "top 500" list will be announced right before SC2003 and discussed in detail at a session of SC2003 on November 18.

Look for another (less speculative) story on Slashdot around then.

The new "top 500" list will be announced right before SC2003 and discussed in detail at a session of SC2003 on November 18.

Look for another (less speculative) story on Slashdot around then.

It's very difficult to make money on software in the Beowulf arena because, duh, it's FREE! You have to make your money on hardware and integration of the hardware and software. Seems that there'll be lots of overhead with all of the developers now on hand at Penguin. Maybe this is why the CEO of Penguin, Marty Sayer, left 2 years ago and is now a VP at AMD.

In addition, for the most part Clustermatic does the same thing and is setup the exact same as a Scyld distribution, granted Scyld does add some neat things of their own. Although Scyld actually has turned a profit here of late, don't get me wrong, I like Becker but I really don't see this one working out in the long run.