mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   NFSNET Discussion (https://www.mersenneforum.org/forumdisplay.php?f=17)
-   -   What kind of computer _is_ colossus? (https://www.mersenneforum.org/showthread.php?t=5494)

fivemack 2006-02-15 09:14

What kind of computer _is_ colossus?
 
Very few of the computer names in the nfsnet participant list give any indication of what the hardware is; I think that information is collected but not given out.

On the other hand, the most obvious feature of the list is cage's computer 'colossus', which has about fifty times the throughput of my G4/1400 Mac Mini, and ten times the throughput of any other computer on the table; it's performing about a third of the searching on its own. :bow:

Is this spare time on a remarkably underused large SMP box, or something more interestingly exotic like a fairly standard PC with lots of memory running lattice rather than line sieving?

On a different note, what kind of extra performance do people with dual-core systems see when running two copies of NFSNET? Sieving is proverbially cache-unfriendly, so I'd fear that the memory bandwidth got constricted.

R.D. Silverman 2006-02-15 13:04

[QUOTE=fivemack]Very few of the computer names in the nfsnet participant list give any indication of what the hardware is; I think that information is collected but not given out.

On the other hand, the most obvious feature of the list is cage's computer 'colossus', which has about fifty times the throughput of my G4/1400 Mac Mini, and ten times the throughput of any other computer on the table; it's performing about a third of the searching on its own. :bow:

Is this spare time on a remarkably underused large SMP box, or something more interestingly exotic like a fairly standard PC with lots of memory running lattice rather than line sieving?

On a different note, what kind of extra performance do people with dual-core systems see when running two copies of NFSNET? Sieving is proverbially cache-unfriendly, so I'd fear that the memory bandwidth got constricted.[/QUOTE]


I get about a 20% throughput increase when 2 copies of my lattice siever
on my (hyper-threaded) 3.4GHz IA-32. (versus running one copy)
Of course one needs enough memory to run 2 copies.

Jushi 2006-02-15 15:24

[QUOTE=fivemack]Is this spare time on a remarkably underused large SMP box, or something more interestingly exotic like a fairly standard PC with lots of memory running lattice rather than line sieving?[/QUOTE]

It's actually a 33-node cluster. They are all Pentium4 3.0GHz machines with 2GB of RAM, running Gentoo Linux. It shows up as one big computer on NFSNET because we use openMosix, which distributes jobs across our cluster, but the jobs still look like they are running on the server node. So NFSNET "sees" a big SMP machine instead of a cluster.

It belongs to the Pure Math department of Ghent University. Currently it's completely idle, but not for long anymore, when I'll start wasting time on the ABC-conjecture.


All times are UTC. The time now is 00:07.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.