Shouldn't the work fetch policy guarantee that each active project always has at least one work unit?

I notice that BOINC 7 tends to fetch a large bunch of work units for Cosmology, which has long running work units. This allows other CPU projects to finish all their WUs, then sit idle for several days before the number of Cosmology WUs declines to the point where anything fetches will happen on the other projects. I don't think this is right.

++PLS

---

Shouldn't the work fetch policy guarantee that each active project always has at least one work unit?

No, not at first. Perhaps later, after you let BOINC do its thing and it's learned the quirks of the projects you've added.

Work is only fetched when the total queue falls down the minimum work buffer value. Then work will be fetched from the project with the highest priority, based on work already done and resource share.

---

Work is only fetched when the total queue [my emphasis] falls down the minimum work buffer value. Then work will be fetched from the project with the highest priority, based on work already done and resource share.

I continue to be unhappy about the fact that there seems to be no clean way of ensuring that each project has some work to do. My own multiproject environment includes CPDN which will always have lots of work because of massive WUs, and so BOINC seems to say to itself that it doesn't need to download WUs from other projects because there's always plenty to do. I would prefer the above algorithm to be implemented on a per project basis.

Also, the lag between WUs finishing and reporting of them puzzles me. I have tried to leave BOINC to its own devices and find that sometimes several days elapse between a bunch of WUs finishing and the completed units disappearing out of my BOINC. When BOINC does finally decide it needs to get more work it generally seems to get lots, because the average work done is so far below the resource usage targets for everyone except CPDN. BOINC runs in feast and famine mode for me.

---

squeak