Too many Gamma-Ray searches
8 Feb 2013 4:22:34 UTC
Topic 196796
(moderation:
I am trying to run Climate Prediction on my CPU and this project on my GPU, plus backing up for when CPDN runs out of tasks (which is often now).
I set CPDN to 300 work share, and this to 1. That should limit it to an occasional CPU task and the GPU which is not used by the other project. I set the cache to .5 days.
Anytime new work for the GPU is downloaded I get gamma ray searches too. I have more than 15 per core, meaning the cache is over a day even ignoring any CPDN work, which there is right now.
Bring the cache down more, wait it out, or is something wrong?
Too many Gamma-Ray searches
)
You could disable downloading of CPU tasks for Einstein@home altogether. Just deselect "Use CPU" in your Einstein@Home preferences. Already downloaded tasks will be processed normally.
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
another option could be to
)
another option could be to reduce the resource share for Einstein to zero, to make Einstein a true backup project.
When I tried running CPDN
)
When I tried running CPDN along side other projects several years ago, boinc ended up face planting in terms of scheduling no matter what I did. At the time it's main failing was seeming not to realize that the hundreds of hours needed to complete the one CPDN WU I was running didn't mean I still had lots of work for my dual core system until my other projects were completely drained; at which point it would download another large lump of tasks and repeat. It sounds like the massive tasks are still throwing it for loop.
This isn't looking good. I
)
This isn't looking good. I changed to a 0 work share.
The GPU only requests the next task when its task is finished. Its fine, though it does waste a minute or so per hour.
Problem is, it keeps saying it is sending two tasks. One being a CPU task. Looking at the log, is there even two tasks? Tasks in progress show two sent at the same time, but this looks like parts of the same unit?
Hi! According to the
)
Hi!
According to the server side scheduler logs, your host is still asking for CPU work, so the E@H server sends some..
Not all tasks require downloading new files. If the required files are already on your host (because they came with a previous task), then you will not see any related downloads in the Manager logs for a new task. The scheduler actually tries to give you tasks that don't require new downloads first before considering other tasks (so called "locality scheduling"). Not all searches support locality scheduling, tho: BRP4 tasks DO require a unique set of data files for each new task.
Cheers
HB
I set the cache to 0 and 0.
)
I set the cache to 0 and 0. Workshare 0.
Boinc is still asking for CPU tasks. The project keeps sending them. I'm up to 90 CPU tasks now. With a single core available (its a backup on the CPU) that is 15 DAYS of cache.
When I used E@H to backup seti on my gpu, Boinc asked for, and received, a single task whenever seti was down. Now, with no other gpu work on the system, its downloading tasks that way. The CPU however is a different beast.
Your preferences must not be
)
Your preferences must not be set right if you are still getting those CPU tasks for GRP's and GW S6's
Go to your account here and reset the "Preferences for this project"
You need to set "Use CPU" to "NO" and then go down the page and at "Run only the selected applications" make sure you only have Binary Radio Pulsar Search (Arecibo):YES box checked and uncheck the other 3 projects.
Then of course click the "Edit Einstein@Home preferences"
The next time after that when you go and update the Boinc Manager for Einstein if you do look at the "event log" it may say it is "Requesting new tasks for CPU and GPU" but in fact it will not be and when you see the tasks that are being downloaded they will only be the GPU tasks for BRP
I change those settings all the time because I still have a couple pc's without a cuda card so I make sure they only get the Gamma Ray tasks and then switch the setting back so my other 4 hosts will only get the BRP cuda's
(tomorrow I am adding a new Geforce card to one of mine so that will mean I only have to change the settings every couple days for just two of my 7 hosts)
Btw you do have 41 GW S6 tasks and 60 GRP tasks in progress right now and just one BRP cuda
http://einsteinathome.org/account/143343
RE: I change those
)
Why aren't you using different locations( home, school, work, standard) for your CUDA and non-CUDA PCs?
SO you wouldn't have to change the settings all the time.
RE: RE: I change those
)
I do that too but I also like to add some GRP's once in a while to go along with the BRP's since they are quad-cores and doing that doesn't slow my BRP's down much but on a 3-core I have I only run the BRP cudas (and T4T X2)
I don't mind doing it that way since I like to mix things up and they are all here in my house.
Sometimes I only run GRP's on those and sometimes just GW's on those 2 hosts.
But then......it is Chris asking for help right now
RE: I set the cache to 0
)
We've just worked out what's going on. It's a bug in the BOINC client.
As Heinz-Bernd said last week, your computer is requesting work for the CPU:
That will be because you have Einstein set as a backup project with resource share 0. At the time of that log, your GPU was idle ("1.00 instances"), and BOINC properly requested work from the backup project.
Unfortunately, because of the bug, every time BOINC requested GPU work, it also requested CPU work as well.
The solution is either to stop doing CPU work at Einstein at all (using the preference controls MAGIC and tolafoph have mentioned), or upgrade to a newer version of BOINC.
From the change log:
â— client: don't request work for backup project for a processor type unless there are idle instances of that type.
That fix was made in v7.0.32, over 6 months ago, but unfortunately the BOINC developers haven't got their act together enough to make any bug fixes into a 'recommended' version since then. I would recommend that you skip all the intermediate versions, and go straight to the current alpha build (v7.0.52) - there are download links at the foot of the most recent entry in that changelog thread.