You're not completely off the mark, but you are a bit off. Speaking from
what I know of the national HPC resources in the US (ALCF, OLCF,
XSEDE), to get on the largest systems you usually have to go through an
application process that shows your programs can scale well enough to
use these large supercomputers, and the schedulers are usually
configured to give preference to larger jobs.
So while it might be true that these systems are seldom used to run a
single job across every single node and processor, efforts are made to
keep out the poseurs and reward those who do use large portions of the
cluster for a single job. In other words, I think your off by a matter
of degree. I think a "small" job on systems like this is probably a few
thousand or tens of thousands of cores.
Hopefully someone from ALCF, OLCF, XSEDE or some other large HPC center
can chime in with more detailed information.
Prentice
On 06/05/2013 11:48 PM, Nathan Moore wrote:
I didn't mean any special insult to the latest top 500 entry. Rather,
that was a comment on the general nature of that list. Does anyone
know, off-hand, how often these big machines run with all compute
nodes dedicated to a single message-passing job? My guess is less
than 1% of the time, and for the most part these machines are used as
batch servers for smaller (N=512, 64, 32, etc) Amber/Gaussian/CFD like
jobs.
Am I far off the mark?
Along similar lines, has Google or Amazon ever published their batch
job capacity (it must be in petaflops...)
NT
On Tue, Jun 4, 2013 at 12:36 AM, Lux, Jim (337C)
<james.p....@jpl.nasa.gov <mailto:james.p....@jpl.nasa.gov>> wrote:
And on SlashDot
http://tech.slashdot.org/story/13/06/03/1231256/full-details-uncovered-on-chinese-tianhe-2-supercomputer
If you want to read a lot of comments from people who don't really
know HPC
From: Nathan Moore <ntmo...@gmail.com <mailto:ntmo...@gmail.com>>
Date: Monday, June 3, 2013 9:55 AM
To: Christopher Samuel <sam...@unimelb.edu.au
<mailto:sam...@unimelb.edu.au>>, "beowulf@beowulf.org
<mailto:beowulf@beowulf.org>" <beowulf@beowulf.org
<mailto:beowulf@beowulf.org>>
Subject: Re: [Beowulf] China to eclipse Titan with 48,000 Intel MICs?
Congratulations on making the world's largest space-heater...
_______________________________________________
Beowulf mailing list, Beowulf@beowulf.org sponsored by Penguin Computing
To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf
_______________________________________________
Beowulf mailing list, Beowulf@beowulf.org sponsored by Penguin Computing
To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf
