Well, I've gotten a few drips and drabs of BRP4 work units a couple of days ago since I discovered the extra preference setting, but no further work for over a day now. Seti@Home and MilkyWay@Home are monopolizing the GPU's right now. Whenever I check the server status page, the results to send for BRP4 is always zero. Are we done with them already? Any new CUDA work in the pipeline? Or do I just drop the project and move on?
Looks like time for more effort towards creating GPU versions of their other applications, then. Not especially important which ones, as long as both the GPU applications versions and enough workunits are available.
When I looked at my preferences, I noticed that there were two lines for BRP, each without a series number. Does this mean to expect BRP5 soon?
On many BOINC projects, lack of some type of workunits often means problems creating workunits of that type that are likely to produce useful outputs. You might want to wait at least a few days before giving up on getting any more GPU workunits. Or do you insist on a continual stream of credits, even for doing nothing useful?
Well, I've gotten a few drips and drabs of BRP4 work units a couple of days ago since I discovered the extra preference setting, but no further work for over a day now. Seti@Home and MilkyWay@Home are monopolizing the GPU's right now. Whenever I check the server status page, the results to send for BRP4 is always zero. Are we done with them already? Any new CUDA work in the pipeline? Or do I just drop the project and move on?
Cheers, Keith
the BRP4 generators are running really slow, and can't meet the demand. Until the hardware is upgraded there're going to be a major shortage of tasks.
Looks like time for more effort towards creating GPU versions of their other applications, then. Not especially important which ones, as long as both the GPU applications versions and enough workunits are available.
Bernd has already hinted that once the Fermi app is stable ( remember it's a completely new search, not just another run), there might be a GPU version for this. The GW search seems to be quite a bit more complex to code for a GPU, tho.
OTOH, there's also work going on to get an OpenCL version of the BRP4 app running, which would, of course, create even more demand for those work units... So the bottom line is that the solution is in the server upgrade (or maybe using GPUs on the server side for work generation, if that isn't done already :-) ). Note that every BRP4 work unit is really a package of 8 sub-workunits, where BRP3 had only 4. That alone gives you an idea of the additional load on the servers.
The lesson I seem to have learned is that with the spotty work generation for BRP4 WU's and the automatically crippled contact to E@H servers every 4 hours just because I run an app_info ........ the better way for me to get work is that whenever I sit down in front of the computers I need to check the server status and see whether any BRP4 tasks are ready to be sent and then do an immediate project update and I have a good chance of snagging some work. Just grabbed 25 WU's for one machine and 17 for the other when I noticed over 700 available.
I sure wish the project administrators would get with the modern BOINC protocol and not impose the 4 hour cripple.
I sure wish the project administrators would get with the modern BOINC protocol and not impose the 4 hour cripple.
You don't want to understand, do you?
Those four hours are imposed as penalty on those crunchers who don't accept the "mandatory" S6 tasks.
Gruß,
Gundolf
Well, I would suggest you go back to the beginning of this thread and read. Your statement is just plain false. Read Bernd's reply to me and his explanation for the message. Nothing at all to do with S6 mandatory work.
This message is a legacy from the times when there was only one application on Einstein@home. The idea is to tell you there is no more work for the application that is configured in the app_info.xml, usually when a new GW run has started.
I'm currently preparing an update of our scheduler, I'll see if I can remove or otherwise fix it.
Couple of points you'll need to bear in mind Keith, as you ponder whether or not to crunch for us :
- GW crunching is mandatory for E@H participation. By preventing CPU work on your machines at a time when we don't have GPU GW apps suitable for release you have subverted this.
- there is no requirement for the project to honor work supply for any arbitrary choice of app_info.xml
- while it may be sad that this does not match your personal expectations, our policy decisions are general and not specific.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Mike, I am happy to work on pulsar searches. I believe every bit of science data gathered is beneficial. I will continue to crunch data for whatever GPU application E@H projects can provide. I find your comment about GW crunching mandatory amusing when the project preferences page itself has the option to not use the CPU which has been set to No for as long as I have attached to the project. Curious as to why that option hasn't been removed if mandatory CPU use is required if attached to the project.
Until the time that the BOINC Manager is completely rewritten to accommodate GPU usage for processing multiple simultaneous tasks, I am forced to use an app_info even though that was not its original intended purpose.
Saw in another thread someone's solution to the automatic 4 hour backoff. Run a CRON job in Linux or Schedule a Task in Windows. I have setup such a task to run and update every hour in the hope that I can continue to grab BRP4 work when it becomes available.
Well, I've gotten a few drips
)
Well, I've gotten a few drips and drabs of BRP4 work units a couple of days ago since I discovered the extra preference setting, but no further work for over a day now. Seti@Home and MilkyWay@Home are monopolizing the GPU's right now. Whenever I check the server status page, the results to send for BRP4 is always zero. Are we done with them already? Any new CUDA work in the pipeline? Or do I just drop the project and move on?
Cheers, Keith
Looks like time for more
)
Looks like time for more effort towards creating GPU versions of their other applications, then. Not especially important which ones, as long as both the GPU applications versions and enough workunits are available.
When I looked at my preferences, I noticed that there were two lines for BRP, each without a series number. Does this mean to expect BRP5 soon?
On many BOINC projects, lack of some type of workunits often means problems creating workunits of that type that are likely to produce useful outputs. You might want to wait at least a few days before giving up on getting any more GPU workunits. Or do you insist on a continual stream of credits, even for doing nothing useful?
RE: Well, I've gotten a few
)
the BRP4 generators are running really slow, and can't meet the demand. Until the hardware is upgraded there're going to be a major shortage of tasks.
http://einsteinathome.org/node/195884
RE: When I looked at my
)
No, the two BRPs are 3 and 4; look at the status page. There're still about 20k WU's that aren't completed yet.
RE: Looks like time for
)
Bernd has already hinted that once the Fermi app is stable ( remember it's a completely new search, not just another run), there might be a GPU version for this. The GW search seems to be quite a bit more complex to code for a GPU, tho.
OTOH, there's also work going on to get an OpenCL version of the BRP4 app running, which would, of course, create even more demand for those work units... So the bottom line is that the solution is in the server upgrade (or maybe using GPUs on the server side for work generation, if that isn't done already :-) ). Note that every BRP4 work unit is really a package of 8 sub-workunits, where BRP3 had only 4. That alone gives you an idea of the additional load on the servers.
HBE
The lesson I seem to have
)
The lesson I seem to have learned is that with the spotty work generation for BRP4 WU's and the automatically crippled contact to E@H servers every 4 hours just because I run an app_info ........ the better way for me to get work is that whenever I sit down in front of the computers I need to check the server status and see whether any BRP4 tasks are ready to be sent and then do an immediate project update and I have a good chance of snagging some work. Just grabbed 25 WU's for one machine and 17 for the other when I noticed over 700 available.
I sure wish the project administrators would get with the modern BOINC protocol and not impose the 4 hour cripple.
Cheers, Keith
RE: I sure wish the project
)
You don't want to understand, do you?
Those four hours are imposed as penalty on those crunchers who don't accept the "mandatory" S6 tasks.
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
RE: RE: I sure wish the
)
Well, I would suggest you go back to the beginning of this thread and read. Your statement is just plain false. Read Bernd's reply to me and his explanation for the message. Nothing at all to do with S6 mandatory work.
http://einsteinathome.org/node/195704&nowrap=true#111130
Couple of points you'll need
)
Couple of points you'll need to bear in mind Keith, as you ponder whether or not to crunch for us :
- GW crunching is mandatory for E@H participation. By preventing CPU work on your machines at a time when we don't have GPU GW apps suitable for release you have subverted this.
- there is no requirement for the project to honor work supply for any arbitrary choice of app_info.xml
- while it may be sad that this does not match your personal expectations, our policy decisions are general and not specific.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Mike, I am happy to work on
)
Mike, I am happy to work on pulsar searches. I believe every bit of science data gathered is beneficial. I will continue to crunch data for whatever GPU application E@H projects can provide. I find your comment about GW crunching mandatory amusing when the project preferences page itself has the option to not use the CPU which has been set to No for as long as I have attached to the project. Curious as to why that option hasn't been removed if mandatory CPU use is required if attached to the project.
Until the time that the BOINC Manager is completely rewritten to accommodate GPU usage for processing multiple simultaneous tasks, I am forced to use an app_info even though that was not its original intended purpose.
Saw in another thread someone's solution to the automatic 4 hour backoff. Run a CRON job in Linux or Schedule a Task in Windows. I have setup such a task to run and update every hour in the hope that I can continue to grab BRP4 work when it becomes available.
Cheers, Keith