Hi Rob, I think you also replied to me when I posted about this on the BOINC boards a few months ago. :-) Thanks for the added info. I'll watch for the Vina 7.03 tasks now if I have problems again. I aborted them all that had been downloaded. All I have now (of the FAAH) are Autodock 7.15 and Vina 7.06. Unfortunately I don't have the time to spare to troubleshoot by selectively aborting some tasks to see if the problem clears, but if it happens again I'll watch to see if it's a 7.03 again. Thanks for your help! :-)

No wings that I can see. A halo out of pipe cleaners maybe. ;-)

Thx NixChix. And thx for the explanation about memory leak. That sounds familiar from wherever I first heard about it. Very annoying, but darn, I was so enjoying blaming it on Vista...

FAHV means Fight Aids a Home VINA

Could help, but not able to make much sense of what you're telling. With LAIM *off* (Leave Application in Memory when [BOINC] paused/suspended), all tasks and their application are supposed to exit memory, and bar memory leaks, free up everything back to the general pool. This is [intentionally] ONLY when a task is past the first checkpoint [job], some able to last 20-30 minutes, maybe longer depending on speed of host [would that be part of what you see?]. Of course, if the setting is to run always, than nothing will unload if the user is at the keyboard or mouse or joyshtick. Make sure to:

1) Set "Run based on preferences", for CPU and GPU, if you need to have free reign.
2) Set memory use by BOINC to default, 50 or 60% when computer is in use, if you allow computing to continue always. This ensures that 1 or more jobs are paused and moved, eventually to VM, when they are big.

AD VINA, VINA being a fast and specializing variation of AutoDock, used by quite a few sciences at WCG and elsewhere, past and present, and thus a fairly well known entity, and VINA meaning nothing, just an out of the air name choice by the programmer, have I lost you already (?), consists of a stager/controller application, not using CPU time but seconds in the hour, and a worker which unloads when each job segment finishes and reloads a new one for the following, as launched by the same stager [it remains in memory for the full duration of a task, unless settings are to unload when paused/suspended. Depending on what the techs decide and the speed of each segment, there can be few, it could be many in a task [I had a few that were probably at the tail end of a target, which just lasted 2-3 minutes for a complete task]. Momentarily all tasks are cut so they don't take too long on an Android device, so there are not many 'jobs' packed in a task, meaning, the whole application including the stager unloads at 100% and a new one [stager] launches with a new task, each in their isolated/sandboxed [shared] memory segment.

How to diagnose... no idea. Some SysInternals tools, world famous amongst the hands-on folk, might do that for you... hit the books on a tool such as Process Explorer. Given that I'm running about 250-275 tasks a day of SN2S and FAHV and not seeing leaking, up time between boots as long as the intervals between MS patch Tuesdays, ergo once a month, I'd consider leaking or lingering as an improbable [Win7-32+64 & Linux-64]. I'd notice.

That said, in case someone gets the urge to bring it up again, batches with 4AOD and 4AOE in the task name of SN2S are large memory usurpers. Yesterday had a bunch running serially, 8 concurrent, biggest 550MB a piece [logged]. Did not notice until I looked in on the machine I 'use'. These 2 come and go and seem to be a thousand, maybe 2 thousand per target... not calcd closely. The bulk of tasks goes between about 50 and 200MB, it varies substantially depending on the search space needed to complete the simulations.

2 cents of inserted guessing.

P.S. What is a historic leak source is the BOINC Manager GUI... some have seen that one go to 1GB. I don't use it, it's not needed, instead use BOINCTasks, a multiclient manager and history tracker of all completed results, on all hosts.

As I mentioned, LAIM Off or On decides if tasks are unloaded when the computer is used, and tasks are passed the first checkpoint, this and of course BOINC running based on preferences, which I read it does since computing does get suspended.

How to control 'total' memory use by BOINC and indirectly also the use of the number of processors on your quad is by setting a low memory % when in use. E.g. suppose your comp has 4GB. Tasks take on average 200 for an overall 800MB. Set memory use while in use to 15% [600MB]. Then 1 or more tasks are paused with "waiting for memory" when big tasks come along. With LAIM on the paused ones moved to VM, one or more cores are freed, and you have those 100% at your disposal... you could even continue, automagically, to BOINC on while working on part of your CPU. More so, there is the feature to make BOINC pause if the CPU load outside of BOINC is greater than an X percent. WCG default is 50%, but if you notice, set it as low as 25% [old default], or lower. Another way to maximizing contribution and you not knowing BOINC is there. You stop 'using' the computer and the 'idle' memory and processors return to full crunching duty.

Finally, as a bonus, if you have specific programs that need more juice or have large disk IO, you can instruct BOINC to suspend, long as these are loaded. Needs just 1 line per application such as <exclusive_app>myworkprogram.exe</exclusive_app>. There's also code to pause GPU computing, e.g. when TD_Scribe is doing his WoT [to pronounce with German accent ;o].