I'm getting only AP WUs now, app_info.xml issue?

Message boards : Number crunching : I'm getting only AP WUs now, app_info.xml issue?
Message board moderation

To post messages, you must log in.

Previous · 1 · 2

AuthorMessage
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14634
Credit: 200,643,578
RAC: 874
United Kingdom
Message 871766 - Posted: 3 Mar 2009, 16:23:25 UTC - in response to Message 871752.  

so the server isn't sending you WU's for your CUDA if you have the AP checked? that would be odd behavior on the server.

It is, but not often enough to keep the balance right.

NB I'm using BOINC v6.4.5 for the time being, and it doesn't ask for separate allocations for CUDA and CPU - it still just asks for one big chunk of work, and accepts whatever mixture the server chooses to send.
ID: 871766 · Report as offensive
Profile Ghery S. Pettit
Avatar

Send message
Joined: 7 Nov 99
Posts: 325
Credit: 28,109,066
RAC: 82
United States
Message 871832 - Posted: 3 Mar 2009, 23:22:01 UTC

Over the past week or so the only WUs sent to my laptop have been AP WUs. No biggie, they take about 3 1/2 days each to run (dual core processor). Fortunately I leave it locked in a cabinet where it can run unmolested at night. And, as others have noted, more credit per hour with AP WUs.
ID: 871832 · Report as offensive
Mike Davis
Volunteer tester

Send message
Joined: 17 May 99
Posts: 240
Credit: 5,402,361
RAC: 0
Isle of Man
Message 871849 - Posted: 4 Mar 2009, 0:18:25 UTC

Ive just left mine off the tick for apv5 but with accept work for other apps ticked. Ive downloaded since doing this, 10 5.03s a few 5.0 and a few hundred mbs. When i have v5 option ticked i got 30 v5s in quick succession and nothing else.
ID: 871849 · Report as offensive
_heinz
Volunteer tester

Send message
Joined: 25 Feb 05
Posts: 744
Credit: 5,539,270
RAC: 0
France
Message 871850 - Posted: 4 Mar 2009, 0:19:24 UTC - in response to Message 871832.  

I'm wondering about your long runtime of 3 1/2 days per ap wu.
I'm running a 5.03 astropulse wu in ca. 12 to 14 hours with a 2GHz Xeon
Try our optimized version from lunatics. http://lunatics.kwsn.net/index.php
happy crunching

D5400XS V8-Xeon
ID: 871850 · Report as offensive
dyeman
Volunteer tester

Send message
Joined: 3 Apr 99
Posts: 23
Credit: 8,493,140
RAC: 0
Australia
Message 871862 - Posted: 4 Mar 2009, 0:55:38 UTC - in response to Message 871766.  

It is, but not often enough to keep the balance right.

NB I'm using BOINC v6.4.5 for the time being, and it doesn't ask for separate allocations for CUDA and CPU - it still just asks for one big chunk of work, and accepts whatever mixture the server chooses to send.


I'm in the same boat. One thing that I've noticed is that with 6.4.5, for me it is only requesting sufficient work for one processor on my 2 dual processor systems (my laptop, and a cheap cruncher box that I just built with a 9800GT). This makes for more micromanagement since I have two single processor machines also. This only started happening when I installed boinc 6.4.5. Is this a "feechur" in 6.4.5 or is there something I've overlooked? Prefs are set to use 100% of the processors.
ID: 871862 · Report as offensive
Andy Williams
Volunteer tester
Avatar

Send message
Joined: 11 May 01
Posts: 187
Credit: 112,464,820
RAC: 0
United States
Message 871880 - Posted: 4 Mar 2009, 1:55:54 UTC - in response to Message 871862.  

The scheduler in 6.4.x is buggy. A very simple case: I am sending this from a dual quad core Xeon machine. Eight real cores. I have "connect every" set to five and "additional work" set to five. There is nothing but AP 5.03 WUs on the machine. Boinc manager is slightly overestimating their time to completion at 17:52:05. Even if they took 24 hours each, 8 cores x 1 WU per core per day x 10 day buffer = 80 WUs on the machine at any given time. Why then are there a total of 35 WUs on the machine, eight of which are >50% complete? Furthermore, if I make a scheduler request by clicking on update, why does the scheduler request zero seconds of work? Rhetorical questions, of course. I have four other eight core systems which show the same behavior.

As a result of this scheduler arithmetic, I ran dry on most such systems during the recent bandwidth saturation, when I was maintaining a more reasonable zero "connect every", three "additional work" buffer

This is a simple case; the scheduler arithmetic on CUDA systems is much more problematic.
--
Classic 82353 WU / 400979 h
ID: 871880 · Report as offensive
Profile KW2E
Avatar

Send message
Joined: 18 May 99
Posts: 346
Credit: 104,396,190
RAC: 34
United States
Message 871900 - Posted: 4 Mar 2009, 2:27:45 UTC - in response to Message 871880.  

Andy:

I have a few dual quad Xeon's with only a few AP WU's as well. Seems to only be keeping few extra WU's even though the cache setting is high.




ID: 871900 · Report as offensive
Cosmic_Ocean
Avatar

Send message
Joined: 23 Dec 00
Posts: 3027
Credit: 13,516,867
RAC: 13
United States
Message 871902 - Posted: 4 Mar 2009, 2:32:46 UTC - in response to Message 871880.  

The scheduler in 6.4.x is buggy. A very simple case: I am sending this from a dual quad core Xeon machine. Eight real cores. I have "connect every" set to five and "additional work" set to five. There is nothing but AP 5.03 WUs on the machine. Boinc manager is slightly overestimating their time to completion at 17:52:05. Even if they took 24 hours each, 8 cores x 1 WU per core per day x 10 day buffer = 80 WUs on the machine at any given time. Why then are there a total of 35 WUs on the machine, eight of which are >50% complete? Furthermore, if I make a scheduler request by clicking on update, why does the scheduler request zero seconds of work? Rhetorical questions, of course. I have four other eight core systems which show the same behavior.

As a result of this scheduler arithmetic, I ran dry on most such systems during the recent bandwidth saturation, when I was maintaining a more reasonable zero "connect every", three "additional work" buffer

This is a simple case; the scheduler arithmetic on CUDA systems is much more problematic.

If you can, or if you feel compelled to, give 6.2.19 a try and see what kind of cache build-up you get. The problem I had with 6.4.5 for the three days I tried it was that it ran my ~4-day cache on a twin-dual rig down to less than one day's worth of work for one CPU/core, and would not ask for more work since tasks were still being run in high-priority. Switching back to 6.2.19 immediately requested ~2.5M seconds of work and filled the cache back up with ~4 days' worth of work...per CPU/core. That was my observation. Besides, no need for >6.2.19 anyway..no CUDA.
Linux laptop:
record uptime: 1511d 20h 19m (ended due to the power brick giving-up)
ID: 871902 · Report as offensive
Andy Williams
Volunteer tester
Avatar

Send message
Joined: 11 May 01
Posts: 187
Credit: 112,464,820
RAC: 0
United States
Message 871912 - Posted: 4 Mar 2009, 2:56:36 UTC

I only used that dual quad Xeon box as an example because it is very simple. In my experience, the scheduler in 6.4.5 - 6.4.6 works in the same (buggy) manner on all machines. I'm used to it now. I can live with it until a version with a smarter scheduler is available.

It's not good for a new or inexperienced user, though. It really only keeps a cache about 30-35% of the target for non-CUDA work. CUDA is much worse.
--
Classic 82353 WU / 400979 h
ID: 871912 · Report as offensive
Profile [B^S] madmac
Volunteer tester
Avatar

Send message
Joined: 9 Feb 04
Posts: 1175
Credit: 4,754,897
RAC: 0
United Kingdom
Message 872032 - Posted: 4 Mar 2009, 14:28:45 UTC

I now how had three AP v5 on the trot and no MBs, just unlucky. Will see what the next one is , if it is an AP will untick the V5 box and run MBs for awhile
ID: 872032 · Report as offensive
Profile Bambi

Send message
Joined: 15 May 99
Posts: 26
Credit: 5,704,701
RAC: 1
United Kingdom
Message 872056 - Posted: 4 Mar 2009, 15:33:41 UTC - in response to Message 871444.  

I've had to juggle between AP and MB on the Setiathome Preferences page in order to keep my CUDA pc stocked with the right amount of AP and MB for CUDA and AK_v8. When I start running low on MB, I just deselect all AP work on the preferences page till I get enough MB, and then reselect AP. When I finish the AP I have, I deselect MB til I get the amount of AP I want, and then allow MB again. After a couple of days I start the process oever again. This is with Boinc 6.4.6 (was the same for 6.4.5).


I have had to do exactly the same thing. When I had no AP work units BOINC downloaded 300+ CUDA WU's and no AP. Now they are done it decided to download 20 AP's and no CUDA.

What has changed? It used to work so well......
Bambi

ID: 872056 · Report as offensive
Previous · 1 · 2

Message boards : Number crunching : I'm getting only AP WUs now, app_info.xml issue?


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.