Somehow the Rakesearch scheduler doesn't work as required:

1. Even when the queue is set to zero, sometimes dozens of tasks are queued up. Result: It is difficult to return any as the first finder (which is the prime goal in mathematics projects).
2. By contrast, at other times, Rakesearch simply doesn't fetch tasks although the server is loaded. I have to manually make it fetch tasks and since most of these are running only a few seconds, it is tedious - to say the least.

I use BOINC v7.20.2 under Win10 with no other project running than Rakesearch (Diagonal Latin Squares App, only).
Maybe you find the time to look into this (idle machines are a bit energy costly these days).

Michael.

---

President of Rechenkraft.net.
Fördern, kooperieren und konstruieren statt fordern, konkurrieren und konsumieren.

Hello Michael!

Any result of current searches (ODLS12 and others) is a small tile in the general mosaic (and this is a problem in the context of issuing badges :D). No any difference whose result will be first.

Thank you for participation! :)

Well, the topic of my question was the scheduler (result discovery just a side remark).

Michael.

---

President of Rechenkraft.net.
Fördern, kooperieren und konstruieren statt fordern, konkurrieren und konsumieren.

I can image some reasons of situations like described, but all of them easily beats by cache for 1 days of computing. And, really, my 10-core Windows VM has never (or very-very rarely) faced with such cases.

No, it is actually a permanent issue and for good reason, I don't want to cache tasks.
Caching will load a max of allowed tasks and if you are "unlucky", these might have quite long runtimes adding up (as, strangely, they appear to be sent out in chunks not randomly - if you expect them to be randomly distributed, as a mathematician maybe some bells should now ring and a broader check initiated for clear statistics) that are unforseeable by BOINC (remember, it believes they all run just a few seconds).
Moreover, if you add a second project, due to the initial short runtime indication of Rakesearch tasks (few seconds each), BOINC has trouble managing proper sequential completion of the queue.
I encountered this strange behavior when combining Rakesearch with Ramanjuan Machine in combination with queuing entirely turned off (although Rakesearch as the single only project still queues up...) - not sure what's going on there...

On top of that and as said above, for unknown reason, Rakesearch sometimes doesn't fetch new tasks for quite some time after it's queue has been completed. You have to manually have it check the server. I could supply screen shots proving it sitting idle as a single hooked-up project for 10 min at least. And I don't like my machine to sit idle as temperature changes are bad for its lifetime. ;-)

Michael.

---

President of Rechenkraft.net.
Fördern, kooperieren und konstruieren statt fordern, konkurrieren und konsumieren.

I can imagine case when internal cache of tasks inside project server instance will be devastated for some time, during usual work. But not for long time. I think that setting tasks buffer size to 0.1 .. 0.2 days may be the best solution.