Check this webpage http://www.tacktech.com/display.cfm?ttid=197

If you're running XP Pro, go to Administrative Services and find where you setup the WCG to run as a service, right click and select properties.
The second tab allows interaction with desktop. HTH

It would be great to have a client that makes a save of the current WU every, say, 10%, so that even that if the WU gets corrupt and aborted, a part of it can be returned nevertheless ?
I've lost a dozen of ultra-long WUs (after more than 20h and as far as 100 hours of crunching, but they would have been shorter and less likely to be corrupt if the air fans had been cleaned up before - I have a notebook P4A 2.6 GHz) due to corruption. That is really counter-productive and annoying.

A simple shutdown & reboot can corrupt the WU, like I saw more than once, and no later than yesterday with a WU that aborted after ~22 hours @ ~80% (WU 2615455). The UD Monitor reports that when the agent is started again, it saves the WU once, and some time later, it aborts.

> Some work units do take much longer (10 times or more) than others.
Yes.
> If you see no progress for many hours, I would use the Task Manager to check for something else which might be consuming CPU time. I have seen virus scan software get stuck in an infinite loop, print spoolers consume 100% of the cpu while waiting for an unconnected printer, and other stranger things.
I know that, but none of those apply here, sorry.

> Rebooting does not currupt a work unit.
Of course, but I saw more than once a WU being corrupted right after (within 1 or 2 checkpoints) the PC is stopped and rebooted (actually, the agent restarted), and that already happened before installing a Linux on my HD. Anyway, my Linux cannot be the cause of corruption: it has read support for NTFS, but not write support.
> Rosetta simply resumes from the last checkpoint. Checkpoints normally occur every several minutes (depending on the speed of the machine) and at most after about an hour (or more on slower machines) if the particular protein fold is non-converging. The progress percentage is updated at the time of the checkpoint.
Yes.

> If you are using UD monitor, what can happen is that a work unit can timeout because it has taken too long (2 weeks run time, 3 weeks wall clock time).
Yes, but this never happened to me, as I always kept a reasonable number of slots (formerly 6, now 4) and never got too many long WUs at a time on the computer.
> If something else modifies the files in any way (including a sector going bad on your disk), the agent simply quits on the current work unit and gets new work.
Well, the disk is scanned from time to time, and it has never had any bad sectors.
> This also happens also if Rosetta crashes for any reason.
Well, it never crashed here in more than 200 WUs. That said, I noticed at least once that a WU aborted a short period of time after another application crashed (that was the application's fault - it was a very buggy beta - not the hardware's).
> Normally, crashes are a sign of some hardware failure or running out of virtual memory. You might want to increase the maximum virtual memory paging file size by 200MB or more to be on the safe side.
I have already run out of VM while WCG was running, due to a buggy GreaseMonkey under Firefox - Firefox had allocated more than 350 MB of VM - but Windows XP smoothly increased the amount of VM, and the WU did not abort.

As far as the entire computer being too hot when saving a WU... maybe. Before I cleaned it up (I made a topic about that, on the Member-to-member forum IIRC), it did overheat all the time, and abort most WUs: the UDMon logs shows it pretty well. When working in a very hot office, it aborted a number WUs the following way: stop the PC at the end of the day, reboot, WU is soon corrupted.

> If you are seeing what you call "work unit corruption," exactly what are the symptoms?
The WU aborts, a directory and a number of files are erased, and the cache slot becomes empty, while there is no seemingly valid reason - especially, no timeout (wall clock time, non-convergence, etc.). I know this is different from a timeout, because I have seen a WU aborting smoothly (hit the maximum time between two checkpoints), and a small result was returned. That WU was definitely too long for my older computer.

I estimate my corrupted WUs are at least three weeks of CPU time total since 2004/12/31, peaking just before I cleaned the computer, which was in dire need of cleaning. If I had had "reliable checkpoints" (what I'm suggesting: checkpoints every 10% or so, which can be sent to the server nevertheless when the WU is corrupted), well, I and WCG would not have lost all that crunching time...
Actually, I did try to restore files from the UDMon backups once, and it worked for some time (the WU could go further than the percentage it aborted at), but it aborted again some time later, and I gave up.
My point is that such a feature could be built in the official software.

I tried again backuping a WU from the UDMon backups. This time, the WU aborted for no reason within seconds after a save, while I was doing nothing fancy with my computer (Explorer, etc.)...
I restored the backup from ~40 minutes earlier. So far, it has crunched more than 4 hours since it started working again on that WU, and is now ~15% further than it was when it aborted (now at ~88%). We'll see if it turns into a returned result. The WU number is 2644630.

PepeG seems to have solved the “Unable to Process Task Data - Backing Off” error at http://www.worldcommunitygrid.org/forums/wcg/viewthread?thread=1858#31030
This problem shows up right after installation. We could add an explanation to the long list under 'Trouble-shooting' at http://www.worldcommunitygrid.org/help/viewTopic.do?shortName=trouble

Looking over this page, the section titled 'Why does my PC show 100% CPU use?' could be followed by one that says 'Why does my PC show 50% CPU use?' that reassures members whose computers use hyperthreading. The next section titled 'My CPU is overheating running while running the agent.' could contain a link to http://www.worldcommunitygrid.org/forums/wcg/viewthread?thread=2683 , the post by Viktors explaining the CPU Throttle Feature. Also, the section title could be changed from 'My CPU is overheating running while running the agent.' to 'My CPU is overheating while running the agent.' without running any risk of confusing the reader.

Added: Troble-shooting is a gigantic page, but if you try to get to it from HELP, it is treated as a title and you can only reach a sub-page. And it can be difficult to get from the sub-page to Trouble-shooting. Try it and see for yourself. This is not acceptable. We need to make it possible for prospective members having trouble to at least reach our help pages. Otherwise they will probably grow too frustrated and give up in disgust.