Only CUDA ABP1 tasks processing - OP is Adhara
13 Dec 2009 1:40:42 UTC
Topic 194671
(moderation:
My application downloaded ABP1 tasks and Hierarchical tasks too.After an ABP1 task is done, another ABP1 is downloaded. BOINC Manager "prefer" to work on CUDA tasks and not on the Hierarchical which are approaching their deadline. How can I stop the manager to download until all tasks are done? I don't want to lose WUs I downloaded .
Thank you !
Cheers,
Gary.
Only CUDA ABP1 tasks processing - OP is Adhara
)
Just click "No new tasks" in the Projects tab of BOINC manager, after highlighting the Einstein@home project.
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
RE: RE: How can I stop
)
Thank you ! It was so simple that I'm ashamed ! :-) I was always looking in the menu !
RE: My application
)
I presume you are talking about this machine's task list where you can see 3 recent APB1 tasks and 4 much older GW tasks which seem to be being ignored.
I'm not sure why the 4 older tasks seem to be left behind but I don't think you need to be too worried yet as none of the tasks are actually in deadline trouble. The deadline is still quite a way off and BOINC should be keeping that under control.
Downloading new tasks is not really causing any problem so I think you can safely 'allow new tasks' again and see what happens. The 4 older tasks will take just over 4 hours each to crunch so at a pinch the whole 4 could be done in less than 10 hours. It would be interesting to see exactly when BOINC (left to its own devices) will decide to crunch them. If they haven't started crunching by say 24 hours before the deadline, you could force the issue by 'suspend'ing whatever new tasks you have in your cache at that point, starting with the very newest, and proceeding until all recent ones are suspended. However I'd be very surprised if that became necessary.
Cheers,
Gary.
Thank you very much, Gary.
)
Thank you very much, Gary. I'll do as you said. It would be interesting to see when Boinc will decide to crunch them. I'm new here, still learning, but thanks to your explanation I understand now better how things work.
Cheers,
Adhara
RE: My application
)
As Gary said, BOINC won't ignore those tasks to the point of missing a deadline, provided they are runnable. In the tasks tab, does it say "Ready to run" in the Status column of these Hierarchical Search tasks? You also appear to have one ABP1 task sent on 4 Dec.
And as you said, BOINC will prefer keeping the GPU busy. In Einstein@Home's case that means keeping a CPU core busy too which can fill the project's resource share. BOINCstats has you attached only to Einstein though, can you confirm you don't have tasks from other projects?
Edit - Oh, and Welcome to the project and the forums!
RE: RE: My application
)
Thanks for correcting my sloppiness :-).
I looked at the older 4 showing in the tasks list and thought I saw 4 GW tasks, but indeed, one of them is an APB1 task. So this is quite puzzling.
@ Adhara - Can you please look at all the 'green' tasks showing on the website which were sent to you on 4 Dec 2009 - particularly the one at the bottom of the second page of the list? Hover your mouse over each taskID so that you can see the name and confirm that you have that exact name on the tasks tab of BOINC Manager. Please let us know if you have all 4 and if they are all listed as either 'Ready to start' or 'Running' in BOINC Manager. I'm beginning to suspect that you must not have all 4 tasks. Perhaps you are missing the single ABP1 task. The server is supposed to 'resend lost tasks' so there must be something unusual going on. It can't even be that the task has become accidently 'suspended' because that would prevent any new task downloads.
This brings me to the point of advising that if you do decide to 'suspend' the individual new tasks to get the old ones to run, you should 'unsuspend' them as soon as the last 'oldie' starts running so that further new downloads can occur when needed.
One other point while I've got your attention. Your list of computers shows three hosts, one of which hasn't communicated since 3 Dec 2009. That machine actually has two outstanding tasks in its list which will expire in about 5 days time. If you still have that computer you might like to switch it on so the tasks can complete. If you don't want to use that machine for crunching, you might like to set 'no new tasks' (NNT) and abort and return the tasks you have so they can be reissued without having to wait to timeout. I'm sure your 'wingman' would appreciate that. A 'wingman' refers to the other person whose computer has been sent the same task for checking and validation of the result. Most people like to see outstanding results get validated quickly and tasks having to timeout and be reissued does slow down that process.
And yes, welcome to the E@H project and please don't hesitate to ask about anything that might be puzzling you.
Cheers,
Gary.
RE: ... I looked at the
)
That ABP1 task was designated for the CPU in the host's very first work request:
The log also had this interesting line (emphasis added):
So BOINC has just one usable CPU core and it's occupied by the GPU task. The scheduler will keep starting CUDA tasks to avoid an idle resource. Hopefully only until the CPU tasks are in deadline trouble.
RE: That ABP1 task was
)
Well spotted!
Yes, I'm sure you are correct.
The only thing left is to ask Adhara if he did intend for BOINC to be limited to just 1 CPU. Of course this is fine but it would lead to the puzzling behaviour that he is seeing.
Cheers,
Gary.
Even to me , my questions
)
Even to me , my questions seem now so naive, because, reading your messages , I understood a lot about how Boinc is working. Here is the story which will make things clear , I hope : On Dec,4 I changed my OS from XP to Win7. My "XP" computer ID was 2187376 and my "Win7" computer ID is 2206256. So, I have 2 computer IDs for the same hardware.Of course, 2187376 is dead now. My Boinc working directory was and is stored on a different partition than the OS, so , after installing Win7, I thought Boinc Manager will continue with the workunits started on XP.Today, I looked for the files - 63288168,63238432 - they are in my Boinc directory, but the manager doesn't see them. Is there any possibility to bring them back to work? I am really sorry for what I did to my wingman!
About those 4 units (3GW +1APB1) : When I ran for the first time on WIn 7,the settings were not to use GPU ,the manager downloaded a few tasks, and later I changed to "Use GPU".Maybe this is the explanation why now my computer is working only on CUDA tasks, while GW + ABP 3.12 are waiting.
Cheers,
"she", Adhara. :-))
RE: Even to me , my
)
Host 2206256 reports as XP Home x86 Edition, Service Pack 2, (05.01.2600.00) whereas 2187376 was XP SP 3. Sorry for a silly question but are you sure the OS where BOINC is currently running is Win7? Also, are the old XP's files p2030_54181_39419_0014_G34.34-01.61.C_2.dm_69 and p2030_54171_45412_0065_G48.16+01.31.C_1.dm_192 in their own einstein.phys.uwm.edu directory separate from the current tasks? If so, they can be retrieved, with some work. Let's worry about that when we know your setup. And don't worry too much about wingmen, lots of tasks expire and einstein only has a 14-day deadline. Of course, always better if you can crunch all that you receive.
In your computing preferences, it should say at 'On multiprocessors, use at most' 100 % of the processors, if you want to use all your cores. If you edit the web preferences, go to the projects tab, select einstein and click update to enforce the changes at your end. If you have edited the preferences in BOINC Manager, those are in force instead.