Page 1 of 1
Determining Processor Info for *nix Boxes
Posted: Wed Sep 03, 2003 5:47 pm
by VLSmooth
Can someone please refresh me on how to determine processor info for *nix boxes? I remember doing so once on the ece machines and wean boxes, determined which were the fast ones, then proceeded to never investigate again.
I originally thought to look under /proc, but all I find there are a bunch of directories with number names.
<excerpt example>
vleeshue@host: ls
0 14879 20568 24861 28358 3113 400 460 6034 890 913
1 14881 206 255 28359 3678 406 4632 6035 897 916
</excerpt>
Posted: Wed Sep 03, 2003 5:57 pm
by VLSmooth
Oh yeah, my goal is to find an ideal server based on server power/online users/processes. When everyone builds, it kinda sucks -_-.
(and no, sampling all servers and conducting empirical tests is out of the question due to the fact I don't want to wait that long)
Posted: Wed Sep 03, 2003 6:26 pm
by Peijen
I thought /proc is processes not processors? But what do I know ...
Posted: Wed Sep 03, 2003 6:35 pm
by Jonathan
look in the directories?
Posted: Wed Sep 03, 2003 6:37 pm
by Jonathan
/proc is lots of interesting little bits about your system.
/proc/cpuinfo is CPU, cache, flags, ID string, stuff like that
/proc/meminfo is memory.
i think the process table is in there, too, but i haven't poked at it.
Posted: Wed Sep 03, 2003 6:54 pm
by VLSmooth
Update: I was lucky enough to have a sysadmin stop by and tell me of a new box that went up ~2 weeks ago. 25 users and running smooth. w00t!
Posted: Wed Sep 03, 2003 7:13 pm
by Jonathan
you should translate this into a clustered batch processing server client program and make everyone use it. like, you know, wossname, condor? only, you know, not sucky.
Posted: Wed Sep 03, 2003 8:36 pm
by VLSmooth
Heh, from what I remember, condor was successful 50% of the time. Luckily, it was successful for my team, very likely due to us getting one sim correct and scripting the rest in perl.
As for batch processing, I'd be surprised if it hasn't been brought up before. In any advent, I'll suggest it (again if need be). Even without a charge number (ie. directly charging a project), it should be easy to show its usefulness.
Posted: Wed Sep 03, 2003 9:10 pm
by Jonathan
what intel does is have a submit script and a daemon. submit script runs in background and submits jobs to daemon. daemon dispatches jobs to machine in the pool. each group is given a number of queue slots. each queue slot has an allocation. pool allocation is enforced in terms of throughput, not active jobs. so if i have 80% of the pool and you have 20% but i don't have any jobs running, you can use all the machines available. later, if i submit some jobs, once yours complete the daemon will only dispatch mine until we reach an 80/20 split or so.