Hi all,

I have run into what I believe should be considered a bug in the BOINC software. I have BOINC running on a 2-core PC under Vista. I an currently running two very long tasks (Primegrid) that will take a total of around 250 hours each to complete. The PC is running about 55% of a day and BOINC has access to all of that time. I am currently about 80 hours into each task and the due date is around Dec.20.

Primegrid is running one of their challenges which runs a bunch of short-term tasks (15-20 minutes each) over 1 day. I was unable to get BOINC to download any of those tasks even by suspending the two long tasks. However, if I RESUME another project, BOINC downloads tasks from that project. It seems to me that I should be able to do the same for the project with the two long tasks.

Any help or explanation will be appreciated.

Thanks,
Jim Wilkins

I an currently running two very long tasks (Primegrid) that will take a total of around 250 hours each to complete. [...]

Primegrid is running one of their challenges which runs a bunch of short-term tasks (15-20 minutes each) over 1 day. I was unable to get BOINC to download any of those tasks even by suspending the two long tasks.

IIRC it is not any unknown bug, the client is rather intentionally refusing to download additional work for the same project, even if it is all suspended. (I do not remember any additional details.)

Peter

Correct. I don't know if it is a bug or not.

Suspended work is still counted as active work, in the work fetch calculations. It's not a bug.

---

But then BOINC should not give preferential treatment to another project, in this case, Yoyo, which was able to download a task, right?

But then BOINC should not give preferential treatment to another project, in this case, Yoyo, which was able to download a task, right?

Why shouldn't BOINC download new work if there's enough time to finish all tasks? Your suspending of the Primegrid task prevented it from downloading new Primegrid tasks though.

You should have upped your wu cache (Maintain enough work for an additional __ days) without suspending any Primegrid tasks.

Gruß,
Gundolf

The distinction is that Jim is considering the two sets of tasks, short and long, as being distinct: and feels that by suspending one, he should be able to get more of the other.

But BOINC sees just one set of tasks, Primegrid, and makes no distinction. By suspending one, BOINC, in its search for another, is forced to look further afield, in this case to Yoyo.

The debate between "multiple projects" and "multiple applications within one project" takes us deep into the design fundamentals of BOINC.

I upped the cache to 10 days while the long tasks were running with no luck.

So I think you are saying that there is some sort of limit applied to all tasks from a given project. If that is correct, how is that limit computed?

So I think you are saying that there is some sort of limit applied to all tasks from a given project. If that is correct, how is that limit computed?

That isn't what I was saying at all.

True, there are mechanisms available to BOINC project administrators to limit the number of jobs you receive at any one time, or the number of tasks you can hold in reserve or in progress on your machine. But those would all imply that your machine was actively asking the project server for more work, and being refused because of one or other of those limits. Those refusals would show up in your message log, and I see no evidence from what you've written so far that that is the case.

Instead, you've told us that you suspended two primegrid tasks. While those two tasks are suspended, you will find that BOINC doesn't request any new work from the primegrid project. And if you don't request work, you'll never receive work - work is never forced on you, by any project.

Perhaps your question would be better directed to the Primegrid administrator, and the way he designs his strategy to get the short tasks returned quickly when there is long-running work already in progress. He or she will have a better idea of the relative need the project has for work of each type.

I'm sorry for being dense here, so please bear with me. Let's say that I had two short term tasks that I suspended because I wanted to run something else from that same project. Would BOINC not request work in that scenario also?

I tried upping the cache to 10 days with the long term tasks executing to no avail. It still would not request new work from Primegrid. However, with the long term Primegrid tasks executing, it did request work from YoYo once I enabled YoYo. That is what is do not understand.

Thanks,
Jim

Let's say that I had two short term tasks that I suspended because I wanted to run something else from that same project. Would BOINC not request work in that scenario also?

Most probably yes. If they all would fit in your cache preference, then pretty sure.

I tried upping the cache to 10 days with the long term tasks executing to no avail. It still would not request new work from Primegrid. However, with the long term Primegrid tasks executing, it did request work from YoYo once I enabled YoYo. That is what is do not understand.

Because, as you described, the PrimeGrid tasks seem to be planned to be complete later than the 10 day crunch cache window. Therefore no more tasks from PrimeGrid.

But as their deadline is far enough, there is enough space to crunch for other projects (e.g. YoYo) alongside, so their download is enabled.

You could try to download and suspend and download and suspend the YoYo tasks, at some point this project's cache should become saturated too.

Peter

You could try to download and suspend and download and suspend the YoYo tasks, ...

I suspect that the aim is NOT to get work from other projects, and that getting some from YoYo was just a test.

Jim
The way that BOINC is designed (at least at present), is that ALL work from A project is counted as work from that project: waiting to run / running / suspended. Just as Richard said.

This issue comes up all the time on cpdn, with people trying to get a different model type by suspending one that they already have. It doesn't work.

If you just HAVE to have some of the shorter WUs before the longer ones complete, (or, at the least, before their approaching completion date takes BOINC into the "Need More Work" zone), then you have 2 choices:
1) Abort one or more of the longer WUs.

2)

Exit from BOINC
Copy ALL of the BOINC (data) folders to another location
Delete all references to the long WUs from the folders, and from client_state.xml, of the original BOINC location
Download short WUs and complete them
Exit from BOINC
Delete ALL of the BOINC (data) folders
Copy back the previously saved copy of the BOINC (data) folders containg the long WUs
Resume crunching the longer WUs

Otherwise, just forget about the short WUs this time around.
Or hope that some will still be available when BOINC needs more work from Primegrid.

You could try to download and suspend and download and suspend the YoYo tasks, ...

I suspect that the aim is NOT to get work from other projects, and that getting some from YoYo was just a test.

Exactly, just a test to confirm that the same should sooner or later happen to YoYo.

Peter

OK...

YoYo was a test, and when BOINC requested YoYo work, that is when I started whining.

I now understand how the cache window is used.

I wanted the shorter tasks because they were part of a challenge. The reason I didn't have this issue before is that I had never exceeded the cache limit.

Thanks for the explanations, folks. I appreciate it.

Jim