Message boards :
Number crunching :
High non-Boinc CPU usage (csrss.exe) while sixtrack WUs processed
Message board moderation
Author | Message |
---|---|
Send message Joined: 28 Sep 04 Posts: 675 Credit: 43,617,129 RAC: 15,853 |
I have Boinc setup to stop processing work if non-Boinc CPU usage goes above 18% (8-core CPU so this means more than one CPU core utilized). I have noticed that with sixtrack WUs running a high CPU usage of csrss.exe (part of Win 7 system Client Server Runtime Process). For each Sixtrack process csrss.exe starts a conhost.exe session and they are the culprit for the additional CPU usage. This results in Boinc starting and pausing the processing in 10-30 second intervals. When processing pauses the non-Boinc CPU usage also goes down and the cycle keeps repeating itself. Is this high activity of csrss.exe a new feature of sixtrack or has it been like this for long? Is this necessary for sixtrack application? I have had this setting in Boinc active now for a few weeks and don't remember if this csrss.exe activity is a longer companion of sixtrack process. But I need this setting as a safeguard against SETI new SoG application crashing the GPU driver, so I don't want to set this limit to 0% which means no limit for non-Boinc CPU utilisation. This didn't come out as coherent as I would have wished but I hope you understand what I mean. |
Send message Joined: 12 Jul 11 Posts: 857 Credit: 1,619,050 RAC: 0 |
Dear Harri, thanks for this input. I am not a Windows expert though. We have certainly not changed SixTrack for a couple of years now (although we have changes coming), I suspect this is a "feature" of Windows 7 (BOINC?). I shall try and get some help and find out what is going on. Eric. (I have a Windows 7 box running SixTrack in my office.) |
Send message Joined: 28 Sep 04 Posts: 675 Credit: 43,617,129 RAC: 15,853 |
I thought that this is not anything new as the sixtrack version number hasn't changed for a while. It just haven't caused any issues when I had the non-Boinc activity setting at 0. When I last time saw this situation, I checked with Sysinternals Process Explorer more details about the process and saw that there was some UserThreads active on the conhost process (don't remember the exact term). What that means I don't know. I just noticed that my laptop (also Win 7 x64) has one sixtrack task running. There I don't see high CPU activity on the csrss.exe. But I caught it when 85% of task was already done. I'll try to monitor if I can see whether this is something only happening on my home computer and whether it is related to only some specific sixtrack tasks or not. |
Send message Joined: 27 Sep 08 Posts: 807 Credit: 651,960,298 RAC: 292,878 |
I have a conhost.exe for every task that boinc runs, from other CERN projects and from WCG, so that is by design in windows. The csrss the link to the x64 part of windows for these x86 applications. If there is a lot of calls through to windows to do actions then the CPU will be higher than when the application is in its internal calculation loops. I imagine as the tasks startup there is a spike and it goes away. |
Send message Joined: 28 Sep 04 Posts: 675 Credit: 43,617,129 RAC: 15,853 |
Thanks for the information. I'll keep a lookout on the matter. I have currently three conhost.exe active under csrss.exe, but that don't seem to match any of the other project tasks running (1 Seti GPU task, 1 Einstein GPU task, 2 CPDN tasks and 4 Seti CPU tasks), only the Seti CPU tasks are using 64 bit applications, others are 32 bit. When a new Seti GPU task started I saw very briefly a fourth conhost.exe on the Process Explorer list. When a new Seti CPU task started I didn't see that or it was so fast that my eyes couldn't register it. |
Send message Joined: 28 Sep 04 Posts: 675 Credit: 43,617,129 RAC: 15,853 |
Today I got four new tasks for my home computer but csrss.exe behaved with its best manner without any noticeable CPU load. I had rebooted the computer last night so maybe that did the trick. |
Send message Joined: 24 Oct 04 Posts: 1127 Credit: 49,747,550 RAC: 10,593 |
Toby is right and also you doing the reboot is usually the first thing to do with any of the Cern project updates or other problems. I found that out when we started T4T/vLHC back in 2011 Rarely if ever had to do that here at Sixtrack though. (surprised you didn't do the free upgrade to Windows 10 on your 3 hosts) Volunteer Mad Scientist For Life |
Send message Joined: 28 Sep 04 Posts: 675 Credit: 43,617,129 RAC: 15,853 |
Yep, I should have thought that immediately but now it looks OK. About updating to Win 10: I try to follow the rule "If it is not broken, don't fix it!" So far I haven't heard anything about Win 10 that would make me to want it. |
Send message Joined: 28 Sep 04 Posts: 675 Credit: 43,617,129 RAC: 15,853 |
Once again I noticed that a conhost.exe process was consuming about half of one CPU core when a sixtrack process started. I was sitting at the computer while this happened. This lasted about a minute and a half and then the CPU load dropped to 0% for this process. While looking at Tthrottle temperature graph and the contents of the stdoutdae.txt for the past 24 hours I see a few incidents when Boinc has stopped and restarted crunching several times in 10 second intervals just after a couple of sixtrack processes were started. The host was rebooted about 23 hours ago so that did not prevent this from happening. So this is a problem to some extent. The good news is that it recovers after the initial phase of the sixtrack calculation is over and continues crunching normally. So far I have not seen any failed WUs because of too many exits. Maybe these computation suspends do not count as exits as the tasks remain in the memory while suspended. |
Send message Joined: 16 Oct 05 Posts: 3 Credit: 100,000,643 RAC: 0 |
Same issue here; I unchecked the "stop tasks above cpu usage..." since my the BM suspended crunching. |
Send message Joined: 28 Sep 04 Posts: 675 Credit: 43,617,129 RAC: 15,853 |
That is a viable solution. I cannot use it because the Seti SoG application tends to crash the GPU driver if my CPU usage goes to 100%. So I have Boinc set to pause crunching if one CPU core (of total 8) gets to full load (CPU usage > 13 %). |
©2024 CERN