| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: BOINC Agent Support
Thread: Projects stay in memory when suspended |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 19
|
|
| Author |
|
|
starhuggrrr
Cruncher Joined: Jun 17, 2006 Post Count: 22 Status: Offline |
I've reported this problem in the BOINC forum previously but the suggestions have not helped.
I have my settings to run projects between certain hours (usually 22:30 to 11:00). I have unchecked Disk & Memory Usage tab > Leave application in memory while suspended, so the projects should not be staying in memory, yet often they do. I'm running BOINC on Vista Home Premium, 32-bit, with 3GB (usable) RAM. I was told in the BOINC forum that I should try editing the configuration file to force a checkpoint early after a project starts running, but this has only eliminated some instances where they would remain in memory. I am still finding my computer suddenly grinding to a halt and when I check the Windows Task Manager, there are project files in memory long after they should have stopped running and been cleared out of RAM. When I mouseover the BOINC icon in the system tray, it shows (correctly) that computing is suspended due to time of day. I don't know if they have been there since the cutoff time or if they have been somehow suddenly loaded into memory. I don't usually have need of checking on this except when my computer suddenly slows to a crawl. This is getting extremely frustrating. I don't have time to spend continually reporting and troubleshooting this bug, which has persisted for months, if not years. I've been participating in WCG for several years, and SETI@home before it. I don't do it for the badges or the stats, but simply to help out the global community projects. I try to maximize the crunching time my computer can give to them. I like being a part of this effort and want to continue, but I'm finding so much of my time and patience wasted over this bug that I'm getting close to considering that I should just drop it, period. I don't want to do that, but the end of my rope is in sight. This bug needs to be fixed. Has anyone else here had this problem? Does anyone know what can be done to fix this? Or who could be contacted who will take action about this? Thanks. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Task certainly remain in memory until first checkpoint is reached, which is intentional. Checkpoints you can't force... that's BS. You can control the minimum interval, so if there's 1 per minute [default BOINC setting is 60 seconds] you can tell, at least the WCG sciences, to not write one sooner than e.g. 5 minutes. Inverse, if a science generates a checkpoint save moment, where the intermediate progress file is smallest, you cant tell it when it does this every 20 minutes, to start writing them every minute.
Whilst, there is some user input detection problems at times under Linux or Windows. For latter it's critical that boinctray.exe is running [the name is duh, for it has nothing to do with the systray icon]. Also, of course, in the activity menu the "run based on preferences" is selected. Also, there's a setting the lets BOINC pause when the user utilization is over an X% of the total CPU load. WCG default setting is 50%, before it was 25%. If BOINC detects this, and run based on prefs is active, this is another control to let BOINC stop while computer is in use. BTW, the 'write to disk at most' which is the checkpoint limiter, does not require file editing. This can be set via the GUI. If you tried this all, try the WCG build 7.2.7... it's an advanced testing beta candidate. It would be a pity to loose you over BOINC not listening to user activity. |
||
|
|
starhuggrrr
Cruncher Joined: Jun 17, 2006 Post Count: 22 Status: Offline |
Thanks Rob, I don't want to have to leave because of this either. I was out for a while tonight and had changed the computing time to run between 11:30 PM until 7:00 PM. Since I got back after 7 but before 11:30, it shouldn't have been running, and yet it was! (not just in memory) WTH?? It really doesn't seem to pay much attention to my settings a lot of the time. :-(
I looked back on the post I made about this in the BOINC forum and it was the cc_config.xml file they had me edit and put in a <checkpoint_debug>1</checkpoint_debug> to monitor if the checkpoints were being made, which they seem to be. (My mistake to think it was forcing a checkpoint.) However, I guess I'm not clear on what checkpoints are and how to gauge whether they will help or hinder my problem. I'm technical enough to think I know what I'm doing sometimes, but not knowledgeable enough to actually know what I'm doing a lot of the time. ;-) Hoping this will help, here are my settings as they have been for a while: PROCESSOR USEAGE TAB: Computing allowed... - While computer is on batteries (CHECKED - it's a desktop, so wouldn't apply anyway) - While computer is in use (CHECKED) - Use GPU... (unchecked) - Only after idle for 5.00 minutes - While processor usage is < 0% - Every day between 22:30 and 11:00 (i.e. from 11:30pm until 11am the next morning; temporarily changed today to 22:30 to 19:00 since I knew I'd be out for a while) - Day-of-week override: none. Other Options... - Switch between appl every 60 min - On multiprocessor systems, use at most 100% (mine is an Intel Quad-Core) - Use at most 90% of CPU tijme NETWORK USAGE TAB: - Min work buffer = 3 days - Max additional buffer - 0.25 days - Everything else is wide open DISK & MEMORY USAGE: Disk usage... - Use at most 10 GB disk space - Leave at least 1 GB disk space free - Use at most 25% of total disk space (harddisk is 465 GB) - Tasks checkpoint to disk every 60 seconds - Use at most 20.00% of page file (swap space) Memory usage... - Use at most 25% when computer is in use - Use at most 100% when computer is idle - Leave appl in memory while suspended (unchecked) Maybe you can see something I could change in my settings? I really would like to get this working properly. I've had this problem for a couple of years now, and every time I get so fed up I'm ready to uninstall it (and have sometimes) I try an update of the program, but it hasn't helped. My current version is 7.0.64(x86), wxWidgets Version 2.8.10. Many thanks for the help. Starhugger |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hi starhuggrrr,
If Task Manager shows 0% CPU for your 4 WCG projects, then they will be loaded into your virtual memory whenever anything else is running in RAM. I think you have narrowed your problem down to an incorrect conclusion and need to look for a different culprit. First, make sure that you have a large virtual memory allocated on your disk. Then find out what is running on the CPU in Task Manager when things grind to a halt. Lawrence |
||
|
|
starhuggrrr
Cruncher Joined: Jun 17, 2006 Post Count: 22 Status: Offline |
Hi Lawrence.
When the tasks are not actively running (Windows Task Manager shows they're not using CPU time, due to preferences), they are still in memory, which slows down my computer when I'm doing memory intensive things (which is often). Since I have the box unchecked for "Leave appl in memory while suspended," they should not remain in memory when they're not active (e.g. during time periods not within the allowable running times); however, they often do remain in RAM. That is my original problem. Then tonight there was the problem where they were actively running (using CPU time, according to Task Mgr) even though it was outside the allowable running period, so that's a second problem. The settings I quoted above show up to 10GB of hard disk memory allocated to this, which is more than enough to run tasks that add up to a few hundred MB at most. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hi starhuggrrr.
You have not said what type of tasks you are trying to run, but I'll have a stab in the dark here and say abort any F.A.A.H. VINA 7.03 tasks you have on board. Going by what you posted here earlier. Quote/ Other Options... - Switch between appl every 60 min - On multiprocessor systems, use at most 100% (mine is an Intel Quad-Core) - Use at most 90% of CPU time /Quote You have a common problem which has happened over the last couple of days with this app. Try it & see if that works. |
||
|
|
NixChix
Veteran Cruncher United States Joined: Apr 29, 2007 Post Count: 1187 Status: Offline Project Badges: 
|
If they are truly suspended the OS memory manager should be rolling the RAM memory image off to hard disk and the tasks should not really be using any RAM until the they are resumed. That is not BOINC client operation; that is done at the OS level.
----------------------------------------Cheers   |
||
|
|
starhuggrrr
Cruncher Joined: Jun 17, 2006 Post Count: 22 Status: Offline |
Hi PPL, Thanks for the suggestion. Yes, all of the tasks I have at the moment are Fight Aids @ Home. I have aborted all the Vina 7.03 now, so we'll see what happens. This is a problem that has gone back over at least a couple of years though, not just a few days. The inconsistency of it might be explained by different types of tasks though. I'll see what happens and let you know. Thanks!
|
||
|
|
starhuggrrr
Cruncher Joined: Jun 17, 2006 Post Count: 22 Status: Offline |
Hi NixChix. Thanks for your reply. If you're right, it sure wouldn't be the first time Vista was the cause of problems. I have found with some other programs that memory often doesn't recycle well (being released and then reused; instead the amount of memory creeps up until I have to reload the whole program). However, if this is a Vista problem then would this not be reported by other Vista users over the years and therefore a well known issue? Have you heard of Vista being the source of memory-related problems for BOINC?
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
This perpetual looping and stuck in memory, going rouge and zombie is far as WCG is concerned a rarity and only surfaced right when FA@H on AD Vina was launched... 2 weeks tops. We hope but have no confirmation other than from observation that v7.06 of this science app fixes all these issues. The workaround was to set the CPU time to 100% as P.P.L. observed.
Waiting with baited breath, yesterday had myself a rouge AD Vina 7.03 [forum reported]... a duo running 3 tasks, and never experienced this before, and this Set and Forget device is running 100% CPU time at that. Two rush jobs had arrived, the 3rd task was a pre-empted FA@H on AD, one in waiting to run state per the GUI and event log, except it was still running per task manager. The CEP2 task, also pre-empted by the rush job had properly paused. Control could not be resumed for the VINA, but strangely aborting via the BOINC manager did work, meaning the client may not have passed the RPC pause instruction properly to the science app. What I did not try was suspend all Ready to Start and then one by one the running tasks to see if that would resume control to the agent, more importantly, started showing the Elapsed time again in normal increment. Can't possible say at this stage if this is an [test] agent v 7.0.67, a science app, or the combination of the 2. Anyway since 7.06 is out of this VINA app the reports have dried up. Otherwise my reading of your posts are that the scheduled crunching based on input times works, bar the bad science app behavior, wrong place wrong time FTM. |
||
|
|
|