Hi Raul,

Unfortunately BOINC sees all WCG sciences as 1 pot, meaning if any task [Beta inclusive] runs for instance twice as long from estimate, BOINC assumes all other tasks to also run twice longer than estimate [your SN2S jobs in this case]. It will normalize again... we all need to allow for BOINC to give it time.

That said, the high variability of WCG tasks, single sciences and across the board, does force to keep the task buffer low as else panic will break out. A 5 days cache suddenly can become 10 days cache by the estimations and then there's deadline trouble. Keep it no higher than 2 days and the issue will rarely crop up.

cheers.

--//--

To add, the servers do know the per-science actual averages, so in principle each science does have it's own FPOPS estimate. Unfortinately the client only keeps 1 duration correction factor [DCF], which you can see in the BOINC Projects view by selecting the WCG line and hitting the properties button on left. That's the multiplicator used to compute the Time To Completion [TTC]

One day, very high on many a wish list, will BOINC get an app-level DCF or something similar. The developer continues to avoid this for unknown reason whilst there has been a trac ticket for years, AND we know a private fork did implement this i.e. it's is doable. Maybe knreed can take some influence and maybe I'll start a discussion at the right time. It's one of the most depressing parts of BOINC... poor time estimates across sciences on the multi-project grid that WCG is.

--//--

? Both are completely divorced issues, andzgrid. WCG knows the app-level down to platform actual run times very well and uses these to cut work to size [were it not that those following the discussions, know all to well that non-deterministic calculations are very difficult to package and cut to a size, so the run times remain within reason for almost everyone.] DCF uses the estimates as were it a 1 science project and even then sciences such as HCMD2 were very hard to size due the enormous variability in computational effort needed even from position to position. Contribution of any smart algorithms to mitigate that are very much welcomed at the developers... open-source, all can contribute that have the skill.

--//--

"My Idea"?

What did I say 3 posts up:

"One day, very high on many a wish list, will BOINC get an app-level DCF or something similar."

It's years and years old the general consensus that it needs addressing, even the developers thinks it's ugly, and as noted, there is a private build that does this. What the mortal side does not know is what's on the roadmap. For one, there's something on the server 700 side that could be deployed to mitigate.

--//--

And so after the client benchmark, the DCF of the client is also ignored when assigning work:

Trust this does carry your approval, andzgrid

--//--