You need to toy a bit and sort the processes by CPU% and CPU Avg and only specify rules for those that use the CPU at a high percent and ignore those that don't. Just fired up my copy 3.99.99 alpha. Not updated for a long time as it does things as I like it... no need to fix things not broken. DON'T use the number of process limiter. That could kill a running task and cause it to be reported as a user abort.

CEP2 number of cached tasks is specified in the website device profile, custom section. It's the only WCG science that allows to specify how many a client can have ''in progress'' and combo'd with PL an ideal companion to manage the individual core loads.

--//--

We just passed the 2.5 million grid results mark (before validation).
http://www.worldcommunitygrid.org/stat/viewProject.do?projectShortName=cep2

PL has load balancing controls out the Ying/Yang so yes, based on CPU loads you can set what should have put the breaks on and even change the affinities when CPU/Mem loads go over definable thresholds ... Task Manager can be binned for good.

In Linux I've integrated another load balancer. Of the 0.2% of the total uptime not going towards crunching, 53% is used to run this kernel patch (not Torvald approved, but it's 5 lines versus his 200 lines that has gone into 2.6.38)

Now it up to you to do the experimenting or move the discussion out of this log. There's a number of recent threads with Linux trickeries to automate affinity and priority, though latter I regard as pointless. BOINC will use 100% of spare cycles anyhow if set to run at max.

Think gb009761 meant that there's also thousands of results that have not been counted yet since stuck in the much maligned PV Jail, waiting on the wingmen. Consider that on average at end of day I've got probably 3 or 4 CEP2 tasks in that state. knreed would know a project average. Maybe the equivalent of 1 production day as a wild guess, maybe less. A very old number at WCG is that over 90% is validated after 4 days from submission. No clue if this improved over the past several years and if this varies per-science, those without redundancy probably having a much smaller PV jail inmateship.

--//--

Dear Aaron, Lawrence, and others,
We had actually ordered business cards with the CEP2 essentials a few days ago and they should be in the mail any day now .

Don’t worry – we have so far only seen problems in a tiny fraction of returned results (a few dozen out of more than a million). Short wus are not uncommon – whenever a calc cannot reach convergence (which just happens sometimes), the consecutive jobs are skipped and the partial wu is returned early. But what you describe wrt wall clock vs cpu time is rather curious.
We btw don’t see data as ‘good’, ‘fair’, ... – it is more about the data being more or less detailed or complete. But here is the thing: We are in principle screening an infinite chemical space (i.e., there are infinitely many potential candidate structures) with a large but ultimately finite resource (i.e., the WCG). So our study is necessarily incomplete and that is perfectly fine. The way we pick our molecules and designed our wus is simply geared towards making the most of the donated computing time within these limitations.

That is destiny telling you to focus on CEP2 – just kidding . But we are glad that Lawrence and SekeRob could clear this up.

Dear martianmoons,
There will be a CEP2 podcast in the near future – we’ll keep you posted on the details.

Best wishes from
Your Harvard CEP team