mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Software

Reply
 
Thread Tools
Old 2012-05-05, 01:19   #1
PageFault
 
PageFault's Avatar
 
Aug 2002
Dawn of the Dead

5×47 Posts
Default P-1, 4GB RAM on 32 bit OS, odd behavior

A friend of mine was about to throw a pile of RAM in the trash, so I stopped him to look at what was available. There were 2 x 1 GB of Corsair DDR2, which is the correct type for my machine. I took them home with me.

System:

Pentium D on Asus P5MT, 2 x 1 GB unknown make > upgrade to 4 x 1 GB.

OS:

Windows Server 2003 Enterprise Ed., DEP and PAE enabled

The OS recognizes all 4 GB of the ram. I looked around on MS site and found some options concerning PAE. 4 GB tuning should be enabled and Address Windowing Extensions. See here

I tried allocating 3 GB as suggested. P-1 ran a minute or so and then restarted after a memory allocation error. I started reducing the allocation and the problem seemed to stop using day / night settings of 2048 MB. Stage 2 finished normally, the next exponent started and once into stage 2 I had a memory allocation error again. So, I backed off a bit, ending at 1920 MB to allow the program overhead to fit into 2048 MB ram.

The client seems to run fine but taskmanager indicates some bizarre behavior. prime95 is indicated at various usage up to 2000 MB but occassionally much less. The client indicates it is using 1880 MB and is running normally.

Overall, the machine is running much smoother, especially during stage 2 (I had 1280 MB allocated before and it was sluggish, with much thrashing of my beloved SCSI disks). I would like to continue running one worker on P-1 but if the results are questionable ...

Any ideas?
PageFault is offline   Reply With Quote
Old 2012-05-05, 02:46   #2
Dubslow
Basketry That Evening!
 
Dubslow's Avatar
 
"Bunslow the Bold"
Jun 2011
40<A<43 -89<O<-88

3·29·83 Posts
Default

As for the much less, Prime95 will be cycling, meaning that every so often throughout stage 2 it's finished with the current block of stuff and has to put the next block into memory; depending on various things, this will result in the total allocation dropping to a few megabytes, then slowly increasing back to the limit over the course of 15-60 seconds (again, depending on various things). The smaller the mem allocation the more this cycling occurs.

For the 2000MB+ vs. 1880 self reported, I've found when I ran P-1 that Prime95 has some memory leaks -- it's never actually using more than it reports, but sometimes at the end of a cycle, when it's loading the next batch it doesn't completely free the previous batch. It's not a serious problem because when Stage 2 overall is finished, all the memory is freed, however during the run those leaks do occur. I've brought them up once or twice, but for various reasons it hasn't been addressed.

(I ran two workers with up to 7200MB total; sometimes the OS would report them using 8,9,10+ GB of mem total. This happened most often when one worker had all 7200 but then the second went into S2; each thread would only report using 3600MB, but the total usage reported by OS jumped into the numbers I mentioned before. It always fixes itself when one worker finishes stage 2, but thrashing happened often enough that I wrote a script to clear out the swap/page whenever it thrashed, otherwise my system would remain almost unusable even after killing Prime95.)
Dubslow is offline   Reply With Quote
Old 2012-05-06, 00:58   #3
PageFault
 
PageFault's Avatar
 
Aug 2002
Dawn of the Dead

5·47 Posts
Default

The cycling was more frequent than per batch change of relative primes. This is new behavior; I am accustomed to keeping taskmanager open to deal with a legacy app that has problems with affinity and this is the first time I saw it.

With server 2003, I have never had any event forcing me to kill an app, or seen evidence of memory leaks - and I do look for those.

Having re-read the stuff on the MS site, I found more details and some things to do. If I can sort it all out, I will ask George to compile a new version for this particular case. All it will take is some header calls ...

I reserved a batch of P-1 tests. If all goes well I will keep one worker on this type. Hopefully I'll snag a few factors ...
PageFault is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Don't understand this mfaktc behavior Chuck GPU Computing 2 2011-06-18 17:24
Question about P-1 behavior JuanTutors Software 0 2006-11-18 02:49
Asymptotic Behavior of a Differential Equation jinydu Miscellaneous Math 7 2006-01-21 20:58
Strange Computer Behavior jinydu Lounge 23 2004-06-08 09:00
Strange behavior on 1.7G Celeron willmore Software 0 2002-09-09 20:17

All times are UTC. The time now is 04:49.

Sun Apr 11 04:49:41 UTC 2021 up 2 days, 23:30, 1 user, load averages: 1.25, 1.32, 1.50

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

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.