"BOINC Client" - keeps increasing memory usage?

[VirtualLarry]

Not the worker threads, but the BOINC client itself, has slowly been creeping upwards in memory usage the last week or two. 890MB so far.

Anyone know why, or if this is normal?

 

[petrusbroder]

Which of the processes do you mean: boinc.exe, boincmgr.exe or boinctray.exe
For me boinc.exe takes 29 852 kB, boincmgr 5 974 kB and boinctray 1 580kB and the size has not changed toady at least.
Please tell more!

 

[Rudy Toody]

Not the worker threads, but the BOINC client itself, has slowly been creeping upwards in memory usage the last week or two. 890MB so far.

Anyone know why, or if this is normal?

Some projects like NFS@Home have memory leaks. Since all apps are spawned from the BOINC client, that memory usage would be attributed to the client.

Reboot to see if the memory usage goes back to normal.

 

[Sunny129]

Hmm...while the issue does sound intuitively strange, i've never noticed (and therefore have never monitored) this kind of behavior before. now that i've looked, it seems that memory consumption by boinc.exe differs drastically across my platforms.

on my AMD X4 925, WinXP 32-bit machine, i run Einstein@Home on 2 CPU cores, and Test4Theory@Home on the other 2 CPU cores. since T4T@H doesn't quite consume a full 2 cores, my CPU usage averages ~90%. boinc.exe consumes ~20MB of system memory on this machine.

on one of my AMD X6 1090T, WinXP 32-bit machines, i run 3 simultaneous Einstein@Home BRP4 tasks on a GTX 560Ti GPU (w/ 1 CPU core allocated to this), and Einstein@Home S6GC tasks on the remaining 5 CPU cores. CPU usage averages ~95% on this machine. boinc.exe consumes ~72MB of system memory on this machine.

on my other AMD X6 1090T, WinXP 32-bit machine, i run 2 simultaneous Milkyway@Home Separation v0.82 tasks on an HD 5870 GPU (w/ 0 CPU cores allocated to this), and Einstein@Home S6GC tasks on all 6 CPU cores. CPU usage is 100% on this machine. boinc.exe consumes a whopping ~530MB of system memory on this machine. while this seems abnormally high compared to my other 2 machines, it has been hovering there for quite some time, and does not appear to be slowly consuming more and more system memory.

 

[Sunny129]

on my other AMD X6 1090T, WinXP 32-bit machine, i run 2 simultaneous Milkyway@Home Separation v0.82 tasks on an HD 5870 GPU (w/ 0 CPU cores allocated to this), and Einstein@Home S6GC tasks on all 6 CPU cores. CPU usage is 100% on this machine. boinc.exe consumes a whopping ~530MB of system memory on this machine. while this seems abnormally high compared to my other 2 machines, it has been hovering there for quite some time, and does not appear to be slowly consuming more and more system memory.

disregard the bolded statement in my previous post above. it turns out that boinc.exe has slowly increased system memory consumption from ~527MB to ~534 over the last 45 minutes...the last time i restarted BOINC was 12 days ago on 2/25. whether boinc.exe system memory consumption has been slowly (but continuously) increasing since then, i don't know. for all i know it could have been oscillating between ~525MB and ~535MB over the past few hours or days. but obviously i'd need to observe for days on end to know for sure, so for now the assumptions are simply based on my immediate observations. instead of waiting a few more days to see what happens, i'm gonna go ahead and restart BOINC now and see if system memory consumption drops significantly. then i'll monitor it to see if it slowly rises over time.

*EDIT* - after restarting BOINC, boinc.exe system memory consumption dropped from 534MB all the way down to 20MB. we'll see where it goes from here...

on a side note, i also checked the system memory consumption of boinc.exe on my other to machines, and they're still at 20MB and 72MB respectively...so these two machines don't seem to have the "boinc.exe increasing memory consumption" issue.

 

[Rattledagger]

Not the worker threads, but the BOINC client itself, has slowly been creeping upwards in memory usage the last week or two. 890MB so far.

Anyone know why, or if this is normal?

Both the v6.10.xx-clients and v6.12.xx-clients got reports of memory-leaks, one memory-leak was fixed in the v6.12.41-client, and this leak would "lose" 120 bytes each time a job started.

So a system running nothing but CPDN would likely not detect this bug, while a multi-GPU-system running 1-minute GPU-tasks would more quickly see the problem.

v6.12.41 isn't on the normal download-page, neither is the last v6.12.xx-client that's v6.12.43. So either download directly from http://boinc.berkeley.edu/dl/ or if you're adventurous you can try the v7-client. Note, you can't downgrade from v7 without losing all work on client.

 

[Sunny129]

Both the v6.10.xx-clients and v6.12.xx-clients got reports of memory-leaks, one memory-leak was fixed in the v6.12.41-client, and this leak would "lose" 120 bytes each time a job started.

So a system running nothing but CPDN would likely not detect this bug, while a multi-GPU-system running 1-minute GPU-tasks would more quickly see the problem.

seems to make good sense to me. one of my machines that doesn't appear to have BOINC memory leaks doesn't do GPU crunching at all, and the other machine that doesn't appear to have a BOINC memory leak completes 3 GPU tasks every ~72 minutes, or 1 GPU task every ~24 minutes. the machine that DOES appear to have a slight BOINC memory leak completes 2 GPU tasks every ~4 minutes, or 1 GPU task every ~2 minutes.

also, thanks for recommending a possible solution - i'll be giving BOINC v6.12.41 or v6.12.43 a try very soon to see if that fixes the leak.

 

[VirtualLarry]

I meant boinc.exe, now it's at 917MB today.

On this machine, I am attached to PrimeGrid (nearly all sub-projects, both CUDA and CPU), Correlizer, World Community Grid (only Clean Energy Project Phase 2), and Neurona@home.

 

[Sunny129]

you know i'm really glad you brought this up Larry, b/c its a problem that i had too, and was completely unaware of it until you said something about it. in the ~6 hours that have passed since restarting BOINC, boinc.exe system memory usage has gone from 20MB to 53MB lol. i'm gonna give BOINC v6.12.41 or 43 a try here shortly and report back before i go to bed...

 

[zzuupp]

Same here!

Could OS flavor be a factor? XP looks relatively unaffected. W7-32 Ouch! W7-64 double Ouch! However, this could also just be the pattern based on relative number of tasks accomplished.

 

[Sunny129]

Same here!

Could OS flavor be a factor? XP looks relatively unaffected. W7-32 Ouch! W7-64 double Ouch! However, this could also just be the pattern based on relative number of tasks accomplished.

no it definitely affects WinXP platforms too - i'm running WinXP 32-bit on the machine that's being affected by the BOINC memory leak...or should i say was being affected by the BOINC memory leak, until i tried Rattledagger's suggestion and updated BOINC v6.12.34 to v6.12.41. its been a few hours now, and boinc.exe system memory usage is still hovering right around the 22MB mark that it was at when i restarted BOINC. i didn't triy BOINC v6.12.43 b/c v6.12.41 worked right away for me.

 

[VirtualLarry]

I restarted BOINC, and boinc.exe went from 935MB to 15MB.

 

[zzuupp]

no it definitely affects WinXP platforms too - i'm running WinXP 32-bit on the machine that's being affected by the BOINC memory leak...or should i say was being affected by the BOINC memory leak, until i tried Rattledagger's suggestion and updated BOINC v6.12.34 to v6.12.41. its been a few hours now, and boinc.exe system memory usage is still hovering right around the 22MB mark that it was at when i restarted BOINC. i didn't triy BOINC v6.12.43 b/c v6.12.41 worked right away for me.

I'm going to monitor for awhile before I do any boinc upgrading.
AFAIK, I haven't had any bad effects despite the memory leak.

 

[Sunny129]

I'm going to monitor for awhile before I do any boinc upgrading.
AFAIK, I haven't had any bad effects despite the memory leak.

i hear ya...aside from v6.12.34's memory leak (which went unnoticed by me for probably 6 months or so), it was an incredibly stable version of BOINC that worked well with my specific hardware configuration and particular mix of projects/applications. for what its worth, i haven't noticed any downsides to v6.12.41 since making the move yesterday. but then again, our hardware profiles and project participation are not identical, so who knows how v6.12.41 might affect you.

so far i'm liking v6.12.41 though...it seems identical to v6.12.41, but without the memory leak. i also want to add that while i thought my Milkyway@Home machine was the only machine significantly vulnerable to the memory leak (due to the fact that it finishes a MW@H task every 2 minutes), it turns out that my Einstein@Home machine is also consuming more and more system memory...its just doing it so slowly that i haven't noticed it until now. while the MW@H machine was leaking almost 10MB/hour (i know! crazy, right?), my Einstein@Home machine is only leaking ~200KB/hour (averaged over ~11 hours). i'm gonna update this machine to v6.12.41 too. this memory leak has never gotten large enough fast enough to cause problems, unlike the MW@H machine...nevertheless, i'm going to update if not for that, for peace of mind...

 

[Sunny129]

well i updated my Einstein@Home machine straight to BOINC v6.12.43, even though v6.12.41 seemed to fix the memory leak on my other machine...i guess i just though i would experiment with both versions so that others could see the possible effects/outcomes. so it turns out that v6.12.43 has not gotten rid of the memory leak...in fact, it appears to have doubled in rate (from 200KB/hour to 400KB/hour) since i updated 12 hours ago. of course a memory leak, whether 200KB/hour or 400KB/hour, is slow enough that it should require days of monitoring for accurate results - time i don't have. in other words, take my memory leak measurements on the Einstein@Home machine with a grain of salt, b/c i only monitored the v6.12.34 leak and the v6.12.43 leak for ~24 hours and ~12 hours respectively.

what makes me somewhat skeptical is that while there does seem to be a slow upward trend in boinc.exe RAM consumption in the short term, it also oscillates up and down a bit in the short term too. so only long term observation is going to reveal the true nature of the memory leak and just how serious it may be. nevertheless, b/c i don't have several days worth of time on my hands to do any such monitoring, i'm going to downgrade the Einstein@Home machine to v6.12.41 later this evening. how do i know it'll be any better than 6.12.43? i don't...boinc.exe RAM consumption on my Milkyway@Home machine (the one that had the monster memory leak before) oscillates just as it does on my Einstein@Home machine, but it seemed to oscillate above and below the same value over a 24-hour period, as opposed to oscillating while slowly increasing in value. so i'm hoping i'll see a similar trend w/ the Einstein@Home machine.

so i'll go to v6.12.41 tonight and report back tomorrow...

 

[Sunny129]

well, as of 10:00pm last night (~24 hours after having switched my Einstein@Home machine to BOINC v6.12.43, and ~12 hours since having taken the first memory leak measurement of 400KB/hour), boinc.exe RAM consumption dropped back down a few megabytes, lending further credence to the idea that this kind of leak must be monitored for well over 24 hours to really get an idea of what's going on. long story short, my assessment that v6.12.43 still had memory leak issues may have been premature, and i see no reason to decrement to v6.12.41 as of yet. i'll keep an eye on it for the next few days, but i have a sneaking suspicion that v6.12.43 will actually prove to be free of memory leaks, and that boinc.exe's RAM consumption oscillation amplitude is simply higher than it is w/ v6.12.41, making it appear in the short term like there actually is a memory leak (if you catch it on the upswing) when there in fact isn't one...