Hi

Your post of the log on [Mar 6, 2013 4:06:32 PM] suggests that BOINC shuts down when CPU load is over 50%. Try 100% (or 95%) and see if that happens.

Re Ubuntu and tracking down excess daemons (Ubuntu-speak: Daemons = background programs in Windows-speak). Look on the forums, and you should see a guide on how to reduce resource drainage.

Try increasing the buffers also - I had some weird results (not downloading many WU, then waiting) on my Windows and Linux installations until I increased the buffers.

Finally, try doing the Memory stress test I suggested:
reboot your machine, when the GRUB bootloader comes up, select MEMTEST and leave it running for 24 hours (or at least overnight). If there are memory glitches, then this should find them. Memory bugs could mean the system is prone to general weirdness. Linux is VERY dependent on your system RAM.

Torchwood 4

How is your "switch applications every XX minutes" setting configured?

The "switch applications every XX minutes" setting is for applications(or work units) running, not different projects. CEP2 work units have a cutoff at 12 CPU hours so it won't take a week to complete, at least not a CPU one. I don't know why you received that CEP2 WU since you claim you selected only HCC in your device profile and AFAIK CEP2 isn't contemplated in the "If no work is available for the projects I selected..." setting in the device profile which you probably selected. So maybe you did something wrong there.

For the general understanding of all readers: 'Swtich Every...' *Only* applies if you have two or more active projects attached to a client, say WCG + FightMalaria@Home. If that is the case, and times up, the switch *normally* only takes place at checkpoint, not earlier. *Only* when there's a rush job in deadline threat will a switch also take place *immediately* to include within the same project, e.g. WCG. At that time and *not* having LAIM on will unload any *pre-empted* task, regarless of whether a checkpoint was made.

Added: Far as I know a task that has not reached it's first checkpoint will *not* be unloaded, regardless if LAIM is on or not.

Not that this does explain anything at all of this thread, but it certainly eliminates !

2 edits: grammar and emphasizing and addition

Here is what I suspect is going on. I have been closely watching my results for the last month. I have not checked clean energy and yet I am getting work units for it. I suspect that these are being executed for a number of hours and than they are being aborted without any benefit to me. I suspect that since the gpu units have started for hcc all other units are being increased in length. I suspect that this is being done to keep the total number of downloads/uploads down to reduce the bandwidth needed for this project. The end result is that I am getting at most 70% of the results I was getting from my computers without a working gpu. The two computers that I do have with working gpu are not being kept up with enough gpu work units so they are spending a lot of time doing non gpu work. I know this because I have a clean energy unit working now and I have seen messages telling me that there are no work units available for my profile(hcc only but will accept others if none are available). I just wish others would go to device installations and compute what they are getting for cpu time/real time.