I recently added the Milky Way project which is supposed to run at 50% of the time with the currently running SETI project, switching every 60 minutes between projects. I received WUs for the CPU, but no GPUs on the first day and nothing since. Any ideas why.
------------------------
07/07/2009 10:03:45 Starting BOINC client version 6.6.36 for windows_intelx86
07/07/2009 10:03:45 log flags: task, file_xfer, sched_ops
07/07/2009 10:03:45 Libraries: libcurl/7.19.4 OpenSSL/0.9.8j zlib/1.2.3
07/07/2009 10:03:45 Data directory: D:\Documents and Settings\Cliff Harding
07/07/2009 10:03:45 Running under account Cliff Harding
07/07/2009 10:03:45 Processor: 4 GenuineIntel Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.40GHz [x86 Family 6 Model 15 Stepping 11]
07/07/2009 10:03:45 Processor features: fpu tsc pae nx sse sse2 pni mmx
07/07/2009 10:03:45 OS: Microsoft Windows Vista: Business x86 Edition, Service Pack 2, (06.00.6002.00)
07/07/2009 10:03:45 Memory: 3.24 GB physical, 6.38 GB virtual
07/07/2009 10:03:45 Disk: 372.60 GB total, 355.86 GB free
07/07/2009 10:03:46 Local time is UTC -4 hours
07/07/2009 10:03:52 CUDA device: GeForce GTS 250 (driver version 18618, compute capability 1.1, 512MB, est. 84GFLOPS)
07/07/2009 10:03:52 Not using a proxy
07/07/2009 10:04:00 Milkyway@home URL: http://milkyway.cs.rpi.edu/milkyway/; Computer ID: 85867; location: (none); project prefs: default
07/07/2009 10:04:00 SETI@home URL: http://setiathome.berkeley.edu/; Computer ID: 4856407; location: home; project prefs: default
07/07/2009 10:04:00 SETI@home General prefs: from SETI@home (last modified 03-Jul-2009 23:37:33)
07/07/2009 10:04:00 SETI@home Computer location: home
07/07/2009 10:04:00 SETI@home General prefs: no separate prefs for home; using your defaults
07/07/2009 10:04:00 Preferences limit memory usage when active to 2989.55MB
07/07/2009 10:04:00 Preferences limit memory usage when idle to 3321.72MB
07/07/2009 10:04:00 Preferences limit disk usage to 300.00GB
07/07/2009 10:04:00 Preferences limit # CPUs to 3
07/07/2009 10:04:01 SETI@home Restarting task ap_18se08aa_B2_P1_00108_20090701_09541.wu_1 using astropulse_v505 version 505
07/07/2009 10:04:01 SETI@home Restarting task 31oc08ac.29555.13979.10.8.48_0 using setiathome_enhanced version 603
07/07/2009 10:04:02 SETI@home Restarting task 13dc08ac.2556.1708.9.8.151_0 using setiathome_enhanced version 603
07/07/2009 10:05:43 Milkyway@home update requested by user
07/07/2009 10:05:46 Milkyway@home Sending scheduler request: Requested by user.
07/07/2009 10:05:46 Milkyway@home Requesting new tasks for CPU and GPU
07/07/2009 10:06:06 Milkyway@home Scheduler request completed: got 0 new tasks
07/07/2009 10:06:06 Milkyway@home Message from server: No work sent
07/07/2009 10:06:06 Milkyway@home Message from server: (won't finish in time) BOINC runs 91.0% of time, computation enabled 100.0% of that

---

I don't buy computers, I build them!

I recently added the Milky Way project which is supposed to run at 50% of the time with the currently running SETI project, switching every 60 minutes between projects. I received WUs for the CPU, but no GPUs on the first day and nothing since. Any ideas why.

First, you didn't get any GPU tasks because the project does not support CUDA at this time... The message from the BOINC client is a generic one. It will make the same GPU request from any project you're attached to, no matter if the project has an application for GPUs or not...

As for the cpu side of things, which is where the "won't finish in time" message came from, right offhand I'd say that the 1250ish tasks that you have assigned to that host over at SETI that are currently "in progress", some of which are due in 4-6 days, is likely the issue...

This should not be a problem, because the SETI WUs will process very quickly. The only ones that will take any amount of time are the Astropulse WUs. The rest CPU WUs usually complete in 2 hours clock time per.

This should not be a problem, because the SETI WUs will process very quickly. The only ones that will take any amount of time are the Astropulse WUs. The rest CPU WUs usually complete in 2 hours clock time per.

3 * 24 * 5 = 360 CPU hours for 5 days at 100% usage for 3 cores

360 / 2 = 180 (maximum tasks able to complete at 100% usage assuming 2 hours each)

number of tasks due within 5 days > 350

There is a huge block of tasks that you requested on June 19th that have deadlines of July 12th, which is 5 days away.

However, the reality is that the tasks there for your CPU are taking 3-4 hours per each... Your system cannot finish the tasks it has on time... Once you clear out some of the SETI tasks, you will be able to get work from here.

Also, if you want to speed up processing with both projects, you might investigate the optimized applications for both projects...

07/07/2009 10:03:45 Starting BOINC client version 6.6.36 for windows_intelx86
07/07/2009 10:06:06 Milkyway@home Message from server: (won't finish in time) BOINC runs 91.0% of time, computation enabled 100.0% of that

Several projects don't like 6.6.36 see http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=952#27182
If you down grade to 6.6.28 you'll get work
If you empty your cache you'll get work

I recently added the Milky Way project which is supposed to run at 50% of the time with the currently running SETI project, switching every 60 minutes between projects. I received WUs for the CPU, but no GPUs on the first day and nothing since. Any ideas why.

First, you didn't get any GPU tasks because the project does not support CUDA at this time... The message from the BOINC client is a generic one. It will make the same GPU request from any project you're attached to, no matter if the project has an application for GPUs or not...

As for the cpu side of things, which is where the "won't finish in time" message came from, right offhand I'd say that the 1250ish tasks that you have assigned to that host over at SETI that are currently "in progress", some of which are due in 4-6 days, is likely the issue...

Brian,

I have the same problem as the original poster: Milkyway requests GPU workunits, which do not exist. Only at the third try it will request CPU workunits. However, I receive the same warning message as mentioned above.

Only when I suspend all other projects and retry for exactly 3 times I will receive CPU workunits - which ofcourse do finish fine within the given deadline.

---

Only when I suspend all other projects and retry for exactly 3 times I will receive CPU workunits - which ofcourse do finish fine within the given deadline.

I think what is confusing both of you is it's not really the MW work that will have a problem, but other work that's on your system. What it's telling you is that due to other work that it has determined it needs to process it won't be able to complete the MW tasks because it needs to focus on the tasks of another project. While your systems could indeed finish 6/core here within 3 days, the other work may miss deadline because of it...that's what it is trying to tell you...

Brian,

the primary problem is, that Milkyway requests GPU-tasks although I neither have the hardware for it nor did I enable GPU-work.
And the error about finishing-time appears only in Milkyway, although I run 6 projects simultanously. I never a saw this message before, both in Milkyway or any other project. It only started with BOINC 6.6.36.

So please don't blame us being too stupid to understand BOINC scheduling when this is obviously a Milkyway-problem!

---

Brian,

the primary problem is, that Milkyway requests GPU-tasks although I neither have the hardware for it nor did I enable GPU-work.
And the error about finishing-time appears only in Milkyway, although I run 6 projects simultanously. I never a saw this message before, both in Milkyway or any other project. It only started with BOINC 6.6.36.

So please don't blame us being too stupid to understand BOINC scheduling when this is obviously a Milkyway-problem!

Work requests are handled by BOINC, not the project application itself. You've been around long enough to know that. You've also been around for the "this project is hogging my cpu" discussion to know about debts and that resource shares are honored over long term rather than second-by-second.

Cliff Harding, the person who started this thread, now has work on his system and he is still using 6.6.36! As I said, once the SETI work cleared, he'd be fine. He had a large batch of work that was due at SETI between the 12th and 13th. Since that time has now passed, voilla...work is downloading from this project more often than just in trickles here and there like it had been. Coincidence????

Please don't blame me for stating that you all may be ignoring certain things in your zeal... As I have said, I do not dispute that there are likely problems with that version, but I think there's too much lumping everything onto that version. When you have a doubling in the runtime with the deadline held constant, you're GOING TO have some of these issues come up. Usually it is from people who are on the bubble. It happened at Einstein a couple of years ago and I asked Bernd to bump the deadlines up to help reduce it. That could be done there, since the research isn't dependent upon the results of tasks before it. Here it is different...

Actually, there was a change in the 6.6.36 client where some some reason, it *the client* wants to request GPU work, even (as in your case) if you don't have a supported GPU -- I've seen this -- and resolved the problem for me by reverting back to 6.4.5 on those workstations which have a supported CUDA GPU and are doing CUDA GPU work (for either GPUGrid or Aqua). If you don't have a CUDA card and are not running Win7 or Vista, you might well find that the 5.4.5 client is a better fit. I know I use that version happily with XP and any non-CUDA workstations. Heck, 5.4.5 will work in conjunction with the 0.19f ATI GPU optimized application here (since that application is working with CPU 'labeled' tasks.

But to repeat, this is NOT a Milkyway problem -- I have seen 6.6.36 request GPU tasks from Climate, POEM, Spinhenge and the like.

So please don't blame us being too stupid to understand BOINC scheduling when this is obviously a Milkyway-problem!

---

Actually, there was a change in the 6.6.36 client where some some reason, it *the client* wants to request GPU work, even (as in your case) if you don't have a supported GPU -- I've seen this -- and resolved the problem for me by reverting back to 6.4.5

So please don't blame us being too stupid to understand BOINC scheduling when this is obviously a Milkyway-problem!

I had forgotten to mention that, so thanks for bringing that point up as well. I was going to ask him to check what happens when work is requested from his other projects as for the "GPU" requests if it was brought up again...

What I'm trying to say is there seems to be a bit of "mob mentality" going on with 6.6.36. I'm sure there are valid issues with it, but piling all issues upon it is probably a bit extreme...

Actually, there was a change in the 6.6.36 client where some some reason, it *the client* wants to request GPU work, even (as in your case) if you don't have a supported GPU -- I've seen this -- and resolved the problem for me by reverting back to 6.4.5 on those workstations which have a supported CUDA GPU and are doing CUDA GPU work (for either GPUGrid or Aqua).

If 6.6.36 is requesting GPU work for a CPU host then it's no wonder that it thinks my CPU system won't finish in time! It also means that the problem won't be seen on projects that don't have GPU apps.

If 6.6.36 IS in fact requesting GPU work on CPU only systems, the fault then lies with 6.6.36 and maybe we should ALL drop our Boinc versions?

Note, this will only solve the "Won't finish in time" issue of not getting work. Other issues may exist which won't be solve by rolling back versions.

Been watching my rolled back host happily crunching MW...

Rod

I've asked DA why, when I have project preferences to not do GPU work and a project may not have a GPU ap, BOINC requests GPU work? His response was....

"...the way the BOINC client finds out that a project doesn't have GPU jobs is by asking it and not getting any. It will keep asking it periodically (since projects may develop GPU apps at any point)."

Could be - then again, I've been less than thrilled with nearly all iterations of the 6.x client. It just seems that with the 6.x client development some things (I'd call them bugs) have gotten less attention in order to implement other features (some of which were not on my list of 'must haves'). Then again, if they added a control feature so that I could suspend network activity at a project specific level -- for those times (like all day every Tuesday for SETI) when projects are not behaving well, now that's a new feature I would like.

What I'm trying to say is there seems to be a bit of "mob mentality" going on with 6.6.36. I'm sure there are valid issues with it, but piling all issues upon it is probably a bit extreme...

---

Interesting -- with projects like SETI having excess I/O pressure, having a client generate more queries.. GPU or CPU or both is something the user can set up at the project level (and once even gets a bit of granularity with that -- using default, Home, School, Work) -- having the client try to bulldoze on its own seems....odd.

Anyway, thanks for that tidbit -- I guess for CUDA GPU projects, I will just sit tight with 6.4.5 (the 6.5 beta works ok for me as well) as it doesn't add that 'feature'.

I've asked DA why, when I have project preferences to not do GPU work and a project may not have a GPU ap, BOINC requests GPU work? His response was....

"...the way the BOINC client finds out that a project doesn't have GPU jobs is by asking it and not getting any. It will keep asking it periodically (since projects may develop GPU apps at any point)."

---

I recive such messages:
18/07/2009 02:01:22 Milkyway@home Sending scheduler request: To fetch work.
18/07/2009 02:01:22 Milkyway@home Requesting new tasks
18/07/2009 02:01:27 Milkyway@home Scheduler request completed: got 0 new tasks
18/07/2009 02:01:27 Milkyway@home Message from server: No work sent
18/07/2009 02:01:27 Milkyway@home Message from server: (won't finish in time) BOINC runs 87.0% of time, computation enabled 100.0% of that

But on this host MW task is completed for ~3 mins, it's HD4870.
And MW project has highest share from all active projects.
Also it has positive STD and zero LTD (other projects have negative debts)

Why server refuses to give work?

I recive such messages:
18/07/2009 02:01:22 Milkyway@home Sending scheduler request: To fetch work.
18/07/2009 02:01:22 Milkyway@home Requesting new tasks
18/07/2009 02:01:27 Milkyway@home Scheduler request completed: got 0 new tasks
18/07/2009 02:01:27 Milkyway@home Message from server: No work sent
18/07/2009 02:01:27 Milkyway@home Message from server: (won't finish in time) BOINC runs 87.0% of time, computation enabled 100.0% of that

But on this host MW task is completed for ~3 mins, it's HD4870.
And MW project has highest share from all active projects.
Also it has positive STD and zero LTD (other projects have negative debts)

Why server refuses to give work?

See this thread.

---

Calm Chaos Forum...Join Calm Chaos Now

Today im getting the same message that cliff got. Im running Seti at 75% and Milkyway at at 25%.
I did run a .25 day cache up until today ,when i backed it off to .10.
I only have 3 Wu from seti that are do the 27 of july but they are shortys 40 mintues. the rest of seti are do Aiug 12. The only thing i can think is that im over my short term debt.
CPU scheduling priority is 0.0
CPU work fetch is 0.0
CPU work fetch deferel is 31;48
CPU work fetch defell interval is 1.04.00
My duration correction factor is 1.8795

Up until today I havent had any problems on this mac getting work. I did install 6.6.36 saturday though.
So am i correct in that im short or long term debt?

---

Up until today I havent had any problems on this mac getting work. I did install 6.6.36 saturday though.

6.6.36 gives a lot of people problems with work.

So am i correct in that im short or long term debt?

You can see the current debt in the Client_state file in Boinc, open file and search for debt or long/short. (not sure where it is at on Macs)

---

Doesn't expecting the unexpected make the unexpected the expected?
If it makes sense, DON'T do it.

Thanks bandit im not sure where it is either. So ill have to poke around a bit.

---