I am trying to run DNETC and Einstein but Einstein is hogging all the resourses, and workunits time out on DNETC. How can I make Einstein play nice with others?
I am trying to run DNETC and Einstein but Einstein is hogging all the resourses, and workunits time out on DNETC. How can I make Einstein play nice with others?
Thanks! Jim
Try not running cpu units at Dnetc and only running gpu units over there. Einstein cannot use your ATI gpu anyway.
... Einstein is hogging all the resourses, and workunits time out on DNETC.
Why do people insist on blaming an individual project for something that is a BOINC responsibility? Ultimately, it is the participant's own choice of preference settings which tell BOINC how to allocate resources to individual projects. If you feel that one particular project is being disadvantaged, you should change resource shares until that project gets the share you would like it to have. Individual projects do not and cannot hog resources all by themselves. A science application cannot just decide to run excessively all by itself. Individual projects will only be run when allowed to by BOINC. BOINC honours your instructions - or at least tries to. Undoubtedly, BOINC has limitations that make it difficult to control all types of project mixes, all of the time. You may need to experiment with BOINC preferences to find the optimal settings that best reflect your wishes. At the end of the day, if you remain unhappy with how BOINC manages individual projects, you need to take your concerns to the BOINC Devs. Individual projects just use BOINC - they don't develop it.
I had a look through your tasks lists on both Einstein and DNETC. At DNETC, I saw about 400+ tasks but I didn't see a single example of a timed out task. All your tasks there seem to be completed quite quickly after arrival. Perhaps they are removed very quickly from the online database. DNETC has a very short deadline period - just 5 days - and you seem to have a large cache setting, judging by how many unstarted tasks you have at both projects. Is that why BOINC is having difficulty getting things right? Perhaps you may need to keep fewer tasks in your cache?
At Einstein, you also have a large cache of work and it looks like tasks aren't being started until they are well on the way to expiry. Judging by the number of tasks (some very close to deadline) that you have aborted, it would appear that you have work cache settings that are far too large for BOINC to properly manage. Rather than continually aborting tasks, why don't you cut down the size of your cache of work so that BOINC doesn't download so many in the first place?
Quote:
How can I make Einstein play nice with others?
By answering the question, "How can I set my BOINC preferences to make it easier for BOINC to run things the way I want them run?" You may need to do quite a bit of experimenting to get close to what you want. Significantly reducing your work cache settings would probably be a good first step.
But the only work units that are running with DNETC are the ones that use my ATI video card that is not supported with Einstein. The ones that would use the CPU never run, or will time out before they do run. Same with any other application I have run. I will ask on the BOINC thread. Sorry to blaim the app so quickly. I don't understand how all this works. I just figured out that my ATI card wasn't supported after much playing. Now I am trying to figure out (on another computer) why after adding an nvidia card, that machine has quit receiving work units for the GPU. Probably a setting I messed with. Anyway, thanks for your help.
But the only work units that are running with DNETC are the ones that use my ATI video card that is not supported with Einstein. The ones that would use the CPU never run ...
Well ... not quite the case :-). I had another look through your tasks list at DNETC and in the most recent 17 pages (340 tasks), there are actually 12 completed CPU tasks. So they are being done although they do seem to take a time to get started. The full story is that there are 54 completed and 200 unstarted GPU tasks together with 12 completed and 73 unstarted CPU tasks. There is also one CPU task that was 'canceled by server'. I think a lot of the problem is being caused by too much work on hand and BOINC is not coping with that.
What cache settings (the first two settings under 'Network usage' on your 'Computing preferences' page) are you using? Have you tried the simple experiment of setting the 'connect to internet' setting to something very low like 0.01 days and the 'extra days' setting to something like 0.2 days and then let BOINC work off all the excess tasks without replacing them? I think you might find that after the excess tasks were gone, BOINC would do a much better job of starting your DNETC CPU tasks in a more timely manner.
Please realise that a large cache setting can lead to two particular problems. Firstly, all the tasks already on hand will delay the starting of more recently received tasks. For a project with a short deadline like 5 days, this can easily lead to deadline misses.
Secondly, in attempting to meet deadlines, BOINC will be forced into running 'at risk' tasks from one project for longer than it normally would, so building up a debt to the other project. At some point it will try to pay back that debt, which in turn leads to the nasty cycle of tasks from the first project not getting started in a timely manner - exactly along the lines you seem to be experiencing. BOINC 'paying back a debt' is often described as some 'nasty project hogging the resources and not playing nice'. Many times this can be solved by reducing the cache settings.
Quote:
... Now I am trying to figure out (on another computer) why after adding an nvidia card, that machine has quit receiving work units for the GPU. Probably a setting I messed with.
I had a look at your list of hosts and I can see a host with an 8800 GTS in it. Is this the one you are referring to? That host has both CPU and GPU tasks (plenty of them) but the problem seems to be that some GPU tasks are completing successfully but quite a few are failing with comp errors. It has received a whole bunch of GPU tasks quite recently. There doesn't seem to be a problem receiving work so I'm not sure why you think there is.
The host with the 8800 is "now" working fine. The failed workunits happened when I added a second GPU, and for some reason, all the workunits that were exising failed. And it took me a couple of days then for me to figure out what happened, and get boinc to start asking for, and einstein to start giving out GPU workunits again. All seems to be working again on that computer.
Also, on the other computer with the ATI HD5870 GPU (which is not compatable with einstein) it IS processing both GPU and CPU workunits on DNETC, and CPU workunits on Einstein. I would prefer that it would not ask for, or process CPU workunits on DNETC, but I have not figured out how to do that . It seemed like when I changed that on DNETC, it also changed it on Einstein.
I've been running this stuff a while, but just started playing with the GPU stuff more, as well as running more than one application.
... I would prefer that it would not ask for, or process CPU workunits on DNETC, but I have not figured out how to do that. It seemed like when I changed that on DNETC, it also changed it on Einstein.
On your account page on the website, there are separate links for 'Computing preferences' and 'Individual project preferences'. The computing preferences are global. The project preferences are specific. You should be able to achieve what you want by appropriate changes in the project preferences. There are settings for 'Use CPU' or 'Use GPU' that you can change.
Be aware that the global changes can be made on the website for any attached project and that the most recent change 'wins'. So you might find it best to make any project specific changes first and then pick a 'master' website on which to make any global changes. The 'master website' should be one that all hosts under your control (if possible) are attached to, so that changes can propagate to all hosts. Think carefully about the implications of making global changes on different websites as you may not quite get what you are expecting :-). You may also need to think about using 'venues' (default, home, school, work) if you want different hosts on the same project to have different settings. Also remember to 'update' particular projects if you want changes in preferences to propagate immediately rather than on the next contact.
Want to run more than one but Einstein a hog
)
Try not running cpu units at Dnetc and only running gpu units over there. Einstein cannot use your ATI gpu anyway.
RE: ... Einstein is hogging
)
Why do people insist on blaming an individual project for something that is a BOINC responsibility? Ultimately, it is the participant's own choice of preference settings which tell BOINC how to allocate resources to individual projects. If you feel that one particular project is being disadvantaged, you should change resource shares until that project gets the share you would like it to have. Individual projects do not and cannot hog resources all by themselves. A science application cannot just decide to run excessively all by itself. Individual projects will only be run when allowed to by BOINC. BOINC honours your instructions - or at least tries to. Undoubtedly, BOINC has limitations that make it difficult to control all types of project mixes, all of the time. You may need to experiment with BOINC preferences to find the optimal settings that best reflect your wishes. At the end of the day, if you remain unhappy with how BOINC manages individual projects, you need to take your concerns to the BOINC Devs. Individual projects just use BOINC - they don't develop it.
I had a look through your tasks lists on both Einstein and DNETC. At DNETC, I saw about 400+ tasks but I didn't see a single example of a timed out task. All your tasks there seem to be completed quite quickly after arrival. Perhaps they are removed very quickly from the online database. DNETC has a very short deadline period - just 5 days - and you seem to have a large cache setting, judging by how many unstarted tasks you have at both projects. Is that why BOINC is having difficulty getting things right? Perhaps you may need to keep fewer tasks in your cache?
At Einstein, you also have a large cache of work and it looks like tasks aren't being started until they are well on the way to expiry. Judging by the number of tasks (some very close to deadline) that you have aborted, it would appear that you have work cache settings that are far too large for BOINC to properly manage. Rather than continually aborting tasks, why don't you cut down the size of your cache of work so that BOINC doesn't download so many in the first place?
By answering the question, "How can I set my BOINC preferences to make it easier for BOINC to run things the way I want them run?" You may need to do quite a bit of experimenting to get close to what you want. Significantly reducing your work cache settings would probably be a good first step.
Cheers,
Gary.
But the only work units that
)
But the only work units that are running with DNETC are the ones that use my ATI video card that is not supported with Einstein. The ones that would use the CPU never run, or will time out before they do run. Same with any other application I have run. I will ask on the BOINC thread. Sorry to blaim the app so quickly. I don't understand how all this works. I just figured out that my ATI card wasn't supported after much playing. Now I am trying to figure out (on another computer) why after adding an nvidia card, that machine has quit receiving work units for the GPU. Probably a setting I messed with. Anyway, thanks for your help.
Jim
RE: But the only work units
)
Well ... not quite the case :-). I had another look through your tasks list at DNETC and in the most recent 17 pages (340 tasks), there are actually 12 completed CPU tasks. So they are being done although they do seem to take a time to get started. The full story is that there are 54 completed and 200 unstarted GPU tasks together with 12 completed and 73 unstarted CPU tasks. There is also one CPU task that was 'canceled by server'. I think a lot of the problem is being caused by too much work on hand and BOINC is not coping with that.
What cache settings (the first two settings under 'Network usage' on your 'Computing preferences' page) are you using? Have you tried the simple experiment of setting the 'connect to internet' setting to something very low like 0.01 days and the 'extra days' setting to something like 0.2 days and then let BOINC work off all the excess tasks without replacing them? I think you might find that after the excess tasks were gone, BOINC would do a much better job of starting your DNETC CPU tasks in a more timely manner.
Please realise that a large cache setting can lead to two particular problems. Firstly, all the tasks already on hand will delay the starting of more recently received tasks. For a project with a short deadline like 5 days, this can easily lead to deadline misses.
Secondly, in attempting to meet deadlines, BOINC will be forced into running 'at risk' tasks from one project for longer than it normally would, so building up a debt to the other project. At some point it will try to pay back that debt, which in turn leads to the nasty cycle of tasks from the first project not getting started in a timely manner - exactly along the lines you seem to be experiencing. BOINC 'paying back a debt' is often described as some 'nasty project hogging the resources and not playing nice'. Many times this can be solved by reducing the cache settings.
I had a look at your list of hosts and I can see a host with an 8800 GTS in it. Is this the one you are referring to? That host has both CPU and GPU tasks (plenty of them) but the problem seems to be that some GPU tasks are completing successfully but quite a few are failing with comp errors. It has received a whole bunch of GPU tasks quite recently. There doesn't seem to be a problem receiving work so I'm not sure why you think there is.
Cheers,
Gary.
The host with the 8800 is
)
The host with the 8800 is "now" working fine. The failed workunits happened when I added a second GPU, and for some reason, all the workunits that were exising failed. And it took me a couple of days then for me to figure out what happened, and get boinc to start asking for, and einstein to start giving out GPU workunits again. All seems to be working again on that computer.
Also, on the other computer with the ATI HD5870 GPU (which is not compatable with einstein) it IS processing both GPU and CPU workunits on DNETC, and CPU workunits on Einstein. I would prefer that it would not ask for, or process CPU workunits on DNETC, but I have not figured out how to do that . It seemed like when I changed that on DNETC, it also changed it on Einstein.
I've been running this stuff a while, but just started playing with the GPU stuff more, as well as running more than one application.
Appreciate your help.
Thanks Jim
RE: ... I would prefer that
)
On your account page on the website, there are separate links for 'Computing preferences' and 'Individual project preferences'. The computing preferences are global. The project preferences are specific. You should be able to achieve what you want by appropriate changes in the project preferences. There are settings for 'Use CPU' or 'Use GPU' that you can change.
Be aware that the global changes can be made on the website for any attached project and that the most recent change 'wins'. So you might find it best to make any project specific changes first and then pick a 'master' website on which to make any global changes. The 'master website' should be one that all hosts under your control (if possible) are attached to, so that changes can propagate to all hosts. Think carefully about the implications of making global changes on different websites as you may not quite get what you are expecting :-). You may also need to think about using 'venues' (default, home, school, work) if you want different hosts on the same project to have different settings. Also remember to 'update' particular projects if you want changes in preferences to propagate immediately rather than on the next contact.
Cheers,
Gary.