Thanks, Sekerob, but Threadmaster is way beyond the simple I'm looking for…

I'd use it myself but I know that some of the people in my team would rather just wait for the new BOINC 'throttle' to come along

Thanks, Kevin, this is very useful

First, remember that UD does use just as much memory. The exact amount of memory used will depend on the project running at that time.

BOINC will unload the memory only depending on the "leave in memory when preempted" setting. I think. I could be wrong, I always leave BOINC running constantly and usually don't even notice it.

jgpastrana, thought i saw an answer to your 'mis-understanding' of what UD agent uses versus BOINC. Let me explain, and this is from a known, not paid by IBM/WCG unbiased perspective @ hand of a screenprint, below, which shows the parts, which I've referenced with line numbers counting from the bottom up:

1. UD.exe (line 5) is the same thing as BOINCmgr.exe (not shown)which uses very little (3mb ram, 65mb Pagefile).
2. A process called UD_xxxxxx.exe (line 4), which is the same as BOINC.exe (not shown), which is the 'intermediate' between the Agent and the Science....uses a few mb.
3. The science itself have varying names. Line 2 is a UD science program, line 3 is a BOINC science program. In this screenprint 'snapshot', the UD science (line 2) uses 282mb Pagefile, but only 39mb RAM. The BOINC science (line 3) uses 29mb RAM and 28mb pagefile....rather the inverse of what u thought, but again this is just a snapshot.

As Didactylos wrote, the pre-empting setting of BOINC determines if at interruption/suspension/snooze/slumber, the science gets unloaded to disk. Certainly, when putting both BOINC and UD in snooze/slumber, most gets unloaded from RAM to disk. ONE exception: In the startup phase of a new Work Unit which can last depending on speed of machine up to 15 minutes or more, the science will remain in RAM! Then u will see up to 285mb of RAM and 285mb of Pagefile. During the process I've even seen momentary peaks of 350mb.

The bad part: If BOINC pre-empting is set to unload to disk (option=NO), chances are, the program looses the time it has crunched from the last checkpoint. Lots of discussion on that subject elsewhere on this forum.

One thing i do not understand is why we keep on hearing about impacts on regular work / production. The science runs at the very lowest priority and always steps aside except for a few minutes when it does disksaving. If your disk is in tip top shape, its hardly noticeble.

PS To talk about strain, don't copy what my HT machine can, but i run both agents simultaneous and I really dont notice except when loading a big app, at which times there is a bit a disk trashing as the science steps aside. If u use Kevin's settings, not even that will occur. RAM is the answer though, 1gb or more ideal. Also notice line 6 bottom up, Threadmaster is the tool i use to manage the CPU allocation. 80% for BOINC, 15% for UD. It's not as hard to implement as people make it out to be (see Treadmaster posts)! Now if i can type as fast to use the spare 5%, i'll be doing a few million words per second ;>)



ERRATA: The GRID.ORG Agent on the face may be the same, but it is in finesse not. The WCG version e.g. uses extra securities e.g. I'm sure u may expect some wind on that revelation