Tom_WCG,

If you'd artificially push part of a job out of RAM into disk based virtual memory, you'll be risking to constantly have jobs swap data from memory to disk and back, making your device slow to very slow.

What you can do is set the memory maximum during use as a percent of total (mine is 75%, of 8GB) and a higher value for when idle (mine is 95% of said 8Gb). Set the percent so that BOINC will run less jobs while you are using the computer [it will suspend them and move them to Virtual memory, if the Leave application in memory while suspended is on]. On a system with 4GB RAM, 4 concurrent CFSW would use 1.2GB maximum, that's 30% [actually that's not much in relative terms to what users use in their applications]. Set memory while using to 20% and just 0.8GB would be allowed, so 2 or 3 jobs will run while you're at the computer, but at maximum performance without swapping back/forth to disk.

And thanks Scribe, but now the Tom_WCG and other readers will know if they do other sciences with high memory use how they could approach this, forcing jobs to pause i.e. use less cores to crunch [temporarily] in an automated fashion.

Tom_WCG, I know how you feel - as, prior to the announcement that the project was coming to a conclusion a lot sooner than originally thought, I was manually allowing 2 CFSW WU's to run (alongside either of the VINA based projects) so as to limit the memory usage.

Now though, I'm just having to 'grin and bear it', allowing 4 CFSW WU's to eat into the 4Gb of memory my lappie has...

Depends on how patient you are. Of course, long as you keep touching that computer, it wont be considered "user idle" until the preference "Only after computer has been idle for xx minutes". Don't know why you chose 40% when in use/not in use, but 100% would certainly send message to go all out and assume you're doing this local. If on the website, of course the client needs to communicate with server before the change takes effect.

The High Priority could hint at high cache or old tasks, so if you have e.g. 7 days cache with 4 cores running, and the client switches to 3 cores, it then has 4/3*7 days buffer, and suddenly you have deadline issues with a near 10 days buffer for 3 cores. [one of the possible explanations]

Solution found! Stupid me!

The problem:
My notebook has 2 cores x 2 hyperthreadings. It should run 4 tasks concurrently. Only 2 was run at at time.

The finding:
While I was fiddling with SekeRob's suggestions above, I mistakenly set xxxx to 50%. My notebook has been running 4 tasks concurrently after I set xxxx to 100%. Am happy crunching now!

[BOING Manager --> Preferences --> Processor Usage --> "On multiprocessor systems, use at most xxxx % of the processors"]