I'm sorry if this has been posted before; however, I didn't find anything directly applicable in my searches (here, google, or otherwise).

I don't understand the "-exit_when_idle" and "-return_results_immediately" flags. Here is what I expect to happen:

I execute the code, download a data package, process the package, upload the results, and quit.

But this doesn't seem to be the case. After the package is processed, the main executable exits but boinc is still running, doing, as far as I can tell, absolutely nothing for all eternity. I have to kill the boinc process and re-execute in order to upload the processed data. And then it downloads a fresh package - which is exactly what I am trying to avoid. Am I missing something here?

I want to be able to define an appropriate walltime for a single job and then submit a batch of jobs via torque/moab to my cluster (just like you used to be able to do with S@H classic). Is this a preference issue, or was boinc somehow not designed to do something like this?

Both are leftover test flags and generally should not be used. Return results immediately has been permanently disabled due the the excessive load it puts on the servers. I am not sure of the status of -exit_when_idle. However if the client can get more work it will not be idle and will not exit.

---

BOINC WIKI

BOINCing since 2002/12/8

Thanks very much for the response. I am glad that you made this very clear.

So, I guess that my next question is how do I tell the client to process exactly 1 data package, ship back the results, and then terminate?

Unless, of course, people are using some other kind of method to batch submit boinc jobs to large HPC clusters... Um, is this the case?

Are you looking to use BOINC locally to manage your own data analysis?

(Sorry if it's a dumb question...)


If that's the case, probably the BOINC_projects mailing list would be the best place to ask.


If not, then I don't have an answer for you.

---

Kathryn :o)

I don't understand why you would want to make BOINC terminate after a single workunit. Just leave it there and it will see there isn't any work available. Otherwise, when you want to release more work (I'm assuming you're talking about your own project), you would have to start BOINC on all machines again!

If what you want is to avoid clients to get more than one workunit, there are server settings for that.

I don't understand why you would want to make BOINC terminate after a single workunit.

Because of the very nature of how HPC clusters work (ie. you share 1000s of CPUs with thousands of other researchers, all waiting to run next in the queue), of course I want it to terminate after a single workunit is processed. I have to get out the way so that other people can run their jobs. That's how scheduling and resource allocation software like Torque and MOAB work.

You simply need to be able to predict how long it takes for a single work unit to process, then you reserve a CPU for approximately this much time. It's as simple as that. I had no problem whatsoever doing this with the classic S@H client. But it seems like maybe this is an impossible task with the BOINC version.

Or maybe I'll just find another way to do this...

If the BOINC core client keeps running, it won't be using CPU. It will just be there waiting for work.

Change BOINC sourcecode if you want it different.

Are you looking to use BOINC locally to manage your own data analysis?

(Sorry if it's a dumb question...)

Not a dumb question at all. I appreciate your response.

I am not actually using BOINC to manage my own data analysis. Rather, I am trying to figure out a method for batch scheduling multiple serial jobs on a HPC linux cluster via TORQUE/MOAB scheduling software.

If the BOINC core client keeps running, it won't be using CPU. It will just be there waiting for work.

Change BOINC sourcecode if you want it different.

No offense, but I think that you are completely missing the point. It doesn't matter to TORQUE/MOAB whether or not I am actually using the CPU. It can't schedule anyone else's job to this node so long as I have reserved it. This is why it's most efficient for me to get in, do my work, upload the results, and then get out of the way, and then wait my turn again for some other node to become available, and then start a brand new process all over again. This is how you share resources in a HPC cluster.

I've searched Google for days now and I am just surprised that nobody has a solution for using BOINC to process serial jobs.

I won't pretend to understand what you're talking about. But, my best suggestion is to ask on the developers email list (see link in previous post).

The people who usually read these boards are pretty knowledgeable about the client software in "normal" (whatever that is...) cases but when interesting questions like yours are asked, a developer is probably the person to talk to.

---

Kathryn :o)

I won't pretend to understand what you're talking about. But, my best suggestion is to ask on the developers email list (see link in previous post).

The people who usually read these boards are pretty knowledgeable about the client software in "normal" (whatever that is...) cases but when interesting questions like yours are asked, a developer is probably the person to talk to.

Thanks for your suggestion. I'll do that.