Seti@Home optimized science apps and information
Optimized Seti@Home apps => Windows => Topic started by: Geek@Play on 14 Jun 2010, 07:38:52 pm
-
Claggy posted on Seti Main this information. http://setiathome.berkeley.edu/forum_thread.php?id=60285&nowrap=true#1004228
quote from his posting.......
The quota should be per app version, 100 for CPU Astropulse_v505, 100 for CPU Seti_enhanced, 100 for Cuda_Fermi, 100 for Cuda32, 100 for Cuda, 100 for ATI Astropulse_v505, etc.
Can someone here verify this statement? Did this information come from Berkeley?
If true, then are the Seti servers going to send the VLAR ONLY to cpu's? (version 603)
-
Did you look at my Beta Application info page i linked to?
Claggy
-
Claggy posted on Seti Main this information. http://setiathome.berkeley.edu/forum_thread.php?id=60285&nowrap=true#1004228
quote from his posting.......
The quota should be per app version, 100 for CPU Astropulse_v505, 100 for CPU Seti_enhanced, 100 for Cuda_Fermi, 100 for Cuda32, 100 for Cuda, 100 for ATI Astropulse_v505, etc.
Yes, that's the plan: you can see it in action at Beta, with the new Application Info (http://setiweb.ssl.berkeley.edu/beta/host_app_versions.php?hostid=28361) pages.
But the implementation at SETI is partial, incomplete, and as yet bug-ridden. (Beta is better, but not yet fully debugged)
If true, then are the Seti servers going to send the VLAR ONLY to cpu's? (version 603)
I doubt it. It could be done, but they haven't shown any enthusiasm to take that up since I first suggested it in February 2009.
If they do decide to implement that, it would mean yet another set of bespoke server code to be written, installed - and debugged: VLARs are not applications (which are handled by BOINC code), but jobs, the exclusive responsibility of SETI. And we know how many spare programmers they have.....
-
As always...........thanks Richard.
-
Does this mean that everybody gets a max of 100 tasks, no matter what system you are running?
I see some reports that there is work distributed for regular Seti tasks but not for optimized versions.
-
I see some reports that there is work distributed for regular Seti tasks but not for optimized versions.
It's untrue, my host recived work and it completely on opt builds of course.
Looks like it even recived 1 AP task for ATI GPU last night.
EDIT: here log of that work request:
15/06/2010 04:01:10 SETI@home Sending scheduler request: To fetch work.
15/06/2010 04:01:10 SETI@home Requesting new tasks for GPU
15/06/2010 04:01:15 SETI@home Scheduler request completed: got 1 new tasks
15/06/2010 04:01:15 SETI@home Message from server: No work can be sent for the applications you have selected
15/06/2010 04:01:15 SETI@home Message from server: No work is available for Astropulse v5
15/06/2010 04:01:15 SETI@home Message from server: Your preferences allow work from applications other than those selected
15/06/2010 04:01:15 SETI@home Message from server: Sending work from other applications
-
I see some reports that there is work distributed for regular Seti tasks but not for optimized versions.
It's untrue, my host recived work and it completely on opt builds of course.
Looks like it even recived 1 AP task for ATI GPU last night.
EDIT: here log of that work request:
15/06/2010 04:01:10 SETI@home Sending scheduler request: To fetch work.
15/06/2010 04:01:10 SETI@home Requesting new tasks for GPU
15/06/2010 04:01:15 SETI@home Scheduler request completed: got 1 new tasks
15/06/2010 04:01:15 SETI@home Message from server: No work can be sent for the applications you have selected
15/06/2010 04:01:15 SETI@home Message from server: No work is available for Astropulse v5
15/06/2010 04:01:15 SETI@home Message from server: Your preferences allow work from applications other than those selected
15/06/2010 04:01:15 SETI@home Message from server: Sending work from other applications
The problem, what I can see is only or mainly with the GPU. I still get some CPU work, got 2x2 AP ;D and some enhanced on another computer.
-
As you can see my example is exactly for asking GPU work.
Maybe you mean problem with asking CUDA-specific work then?
-
As you can see my example is exactly for asking GPU work.
Maybe you mean problem with asking CUDA-specific work then?
Sorry, that's what I mean. But what I understand that every plann class now has it's own work assigned to it. Like cuda 2.3, cuda version x, cuda version y.
Just like the version numbers now.
-
Maybe, but my observation shows that it's not so clear.
ATI GPU app is definitely separate app, it recived only 1 task but not same quota limiting message appears for GPU work requests too...
-
Maybe, but my observation shows that it's not so clear.
ATI GPU app is definitely separate app, it recived only 1 task but not same quota limiting message appears for GPU work requests too...
The quota limit message clearly doesn't appear until the servers think you've downloaded that many in the current day for a specific application. There's a very strong suspicion that the code which resets the count of downloads isn't being called for anonymous platform. That is, the limit may not be 100 per day but rather 100 since the server changes were put in place. My Pentium-M has gotten 15 MB and 1 AP in the last week, my P4 even fewer, so I can't judge directly.
Joe
-
I don't know if this will make any sense but I've not got any of these strange messages on my E5400 with a 9500GT running Vista Home Premium 32bit. I keep a three day cache and only have around 120 or so work units on my machine, 52 CPU and 70 GPU tasks according to the rescheduler. I have no problem keeping the tasks topped off. Finish a couple, report and get some more. Is it because I keep my requests below the 100 limit that it keeps me filled up with work?
-
I don't know if this will make any sense but I've not got any of these strange messages on my E5400 with a 9500GT running Vista Home Premium 32bit. I keep a three day cache and only have around 120 or so work units on my machine, 52 CPU and 70 GPU tasks according to the rescheduler. I have no problem keeping the tasks topped off. Finish a couple, report and get some more. Is it because I keep my requests below the 100 limit that it keeps me filled up with work?
The less you ask the more you get. ;D Just got one Cuda on my laptop, just seconds for the weekly shutdown.
Hmm, it's exactly the same 64 bit Lunatics installer as the other machines that need a lot more and get nothing.
-
There's a very strong suspicion that the code which resets the count of downloads isn't being called for anonymous platform. That is, the limit may not be 100 per day but rather 100 since the server changes were put in place.
Joe
I'm not seeing this. Since this saga began, my pc downloads 100 or so workunits every day and then stops with quota reached.
-
I'm not seeing this. Since this saga began, my pc downloads 100 or so workunits every day and then stops with quota reached.
That's interesting. What version, exactly, is it downloading for? (stock/anon, OS, platform, bits - anything you can thnk of).
-
Sorry, I've posted a Out Off Topic ::)
My WIN XP64 CUDA (2x GTS250) host received only a few, probably Einstein was ahead, LTD, IMO.
And my ATI host received lots of CUDA compute capable MB WU's, which are now done by Q6600 CPU.
But the host did receive ~40, maybe some more, AP WU's, which is absolutely fabulous, for testing and
processing these, cause SETI MB (Optimized SSSE3), Docking, MW and Leiden, are running together.
Collatz C. is still on n.n.tasks.
Will do a test, offline, with a few AP & CC WU's.
AP WU's are running OK, taking 2.5 - 5.5 hours runtime.
-
I'm not seeing this. Since this saga began, my pc downloads 100 or so workunits every day and then stops with quota reached.
That's interesting. What version, exactly, is it downloading for? (stock/anon, OS, platform, bits - anything you can thnk of).
I see this on my linux machine with AKv8 and crunc3r's cuda app. Of course fully anon platform. And I think I see this also on my other winxp machine with gtx470 running jason's x19, again fully anon platform. And I get, in both of them, the stupid message about non usable app_info.xml.
Another strange thing is that the message says about a quota of 100 tasks while on Berkeley's pages the linux machine says a quota of 64/day and the winxp machine 97/day.
-
I see this on my linux machine with AKv8 and crunc3r's cuda app. Of course fully anon platform. And I think I see this also on my other winxp machine with gtx470 running jason's x19, again fully anon platform. And I get, in both of them, the stupid message about non usable app_info.xml.
I've got no point of comparison for the Linux platform. But the WinXP, if confirmed, would contradict my observation, and hence destroy the hypothesis. Could you possibly confirm when you first reached the '100 quota' message?
Another strange thing is that the message says about a quota of 100 tasks while on Berkeley's pages the linux machine says a quota of 64/day and the winxp machine 97/day.
This one, at least, is easy. Beta also showed the obsolete version of the quota until I complained to DA, and he (very quickly) knocked up the replacement 'Application Info' pages for Beta. They just haven't migrated across to the main project yet - unless they appear after the outage.
-
I've got no point of comparison for the Linux platform. But the WinXP, if confirmed, would contradict my observation, and hence destroy the hypothesis. Could you possibly confirm when you first reached the '100 quota' message?
I don't remember exactly, I guess the first day the crappy server code was on? A gtx 470 burns more than 100 workunits a day.
-
This picture is from today. It got 100 tasks and then the usual, by now, reached 100 tasks quota message.
-
This picture is from today. It got 100 tasks and then the usual, by now, reached 100 tasks quota message.
What's more interesting is that it got the '100 tasks quota' message at 09:58:57, and then got new tasks at 10:00:27.
What's your time-zone - what's that in UTC?
-
It's UTC+3 You'll also see that before 10:00 it says quota of 85 tasks. The picture doesn't show it but after the downloads the quota went back to 100.
-
Looks like there's been further changes: Changeset 21753 (http://boinc.berkeley.edu/trac/changeset/21753)
Timestamp: 06/15/10 15:21:57 (less than one hour ago)
Author: davea
Message: - scheduler: restore scaling of daily quota by # processors
and/or config.gpu_multiplier
- client: msg tweak
Claggy
-
Ah - "[version] [AV#%d] daily quota exceeded\n", av.id
That'll be useful.
-
Ah - "[version] [AV#%d] daily quota exceeded\n", av.id
That'll be useful.
Not to us, that's the server log and unchanged in that regard. The change should avoid the "reached max..." message in most cases, but it'll be the unscaled max_jobs_per_day when shown.
Joe
-
Not to us, that's the server log and unchanged in that regard. The change should avoid the "reached max..." message in most cases, but it'll be the unscaled max_jobs_per_day when shown.
Joe
When taking GPU's into Quota-##, is an average taken, since the big differences between 'older' and 'latest' or FERMI GPU's capabilities and Compute Power. (ATI-GPU's, too ofcoarse)
Or by the (G)FLOPS ##, estimated by BOINC (6.10..56)?
-
When taking GPU's into Quota-##, is an average taken, since the big differences between 'older' and 'latest' or FERMI GPU's capabilities and Compute Power. (ATI-GPU's, too ofcoarse)
Or by the (G)FLOPS ##, estimated by BOINC (6.10..56)?
Quota is meant to protect the project from a host which has gone bad, and is the same no matter what the speed of the crunching resource. The slowest CPUs with about 40 Whetstone MIPS have the same 100 basic quota as the fastest. Same goes for GPUs, with the change yesterday they should be back to all having a basic quota of 500. Even the slowest GPUs can trash that many very quickly if they go bad.
The project's config.xml specifies the 100 basic quota and the 5 multiplier for GPUs. They don't want hosts to run out of work if the host is producing good results, if necessary they may increase those settings. But they won't set them so high that the protection is ineffective.
Joe