| Author |
Message |
|---|
 warddr Volunteer tester
Send message
Joined: 7 Feb 08
Posts: 254
ID: 18735
Credit: 10,383,352
RAC: 0
          
|
|
There is something wrong with some recent PPS wu's, all computation errors:
http://www.primegrid.com/workunit.php?wuid=133244468
http://www.primegrid.com/workunit.php?wuid=133014013
Any idea what's going on?
It are all linux systems running boinc 6.10.17
____________
 |
|
|
|
|
There is something wrong with some recent PPS wu's, all computation errors:
http://www.primegrid.com/workunit.php?wuid=133244468
http://www.primegrid.com/workunit.php?wuid=133014013
Any idea what's going on? [/quote]
We updated LLR and missed to change the size.
You will see that the new LLR is much faster.
You should get working WU's now.
Lennart
|
|
|
warddrVolunteer tester
Send message
Joined: 7 Feb 08
Posts: 254
ID: 18735
Credit: 10,383,352
RAC: 0
|
|
It's getting worse:
http://www.primegrid.com/results.php?userid=18735&offset=0&show_names=0&state=5&appid=
____________
|
|
|
|
|
It's getting worse:
http://www.primegrid.com/results.php?userid=18735&offset=0&show_names=0&state=5&appid=
We need some Windows user to clean up all old wu :)
It will be better.
Lennart |
|
|
|
|
|
I've got 10 cores on PPS and SGS (but the server gives out far more PPS) so that should help :)
On a related thought, I've not looked at LLR versions, but are there possible speed gains for Windows users, and/or other LLR based projects too? |
|
|
|
|
I've got 10 cores on PPS and SGS (but the server gives out far more PPS) so that should help :)
On a related thought, I've not looked at LLR versions, but are there possible speed gains for Windows users, and/or other LLR based projects too?
The speed gains are the result of (let me quote from the whatsnew.txt file of Prime95 v26.2):
- Faster FFTs for most SSE2-capable CPU architectures.
in the gwnum-lib that is also used by the LLR application.
So the speed gains should occur on every platform* and many LLR sub projects. The speed gains may vary depending on the CPU you use for primality testing and the subproject it crunches on, in particular the current FFT length it uses.
*as soon as the new binaries are released
____________
|
|
|
warddrVolunteer tester
Send message
Joined: 7 Feb 08
Posts: 254
ID: 18735
Credit: 10,383,352
RAC: 0
|
|
For the new wu's, the speed-up is great:
from about 12 minutes to 8 minutes, that's 33% speed gain.
____________
|
|
|
|
|
|
that's nice!
not nice is this:
Exit status -177 (0xffffffffffffff4f)
after some minutes of runtime!
on hosts which formerly did not have any problems... |
|
|
Ken_g6Volunteer developer
Send message
Joined: 4 Jul 06
Posts: 915
ID: 3110
Credit: 183,164,814
RAC: 0
                       
|
|
I was noticing, some of those errors are "Maximum disk usage exceeded" errors. Perhaps disk space allowance needs to be increased? Or could there be a disk space leak or something?
____________
|
|
|
VatoVolunteer tester
Send message
Joined: 2 Feb 08
Posts: 785
ID: 18447
Credit: 263,436,450
RAC: 0
             
|
|
There was an error relating to that, which was fixed, but there are still some faulty WUs out there.
It'll disappear eventually afaik, but it's definitely not a local client-side issue.
____________
 |
|
|
samuel7Volunteer tester
Send message
Joined: 1 May 09
Posts: 89
ID: 39425
Credit: 257,416,435
RAC: 0
            
|
|
If people want to avoid making errors on the reissues, there is a way to fix the WUs client-side.
The following is intended for advanced users. No responsibility is assumed by me for the loss of cache or any other problems caused by your attempts at the fix. I recommend suspending network activity and making a backup of the whole data directory tree prior to editing client_state for easy restoration of a good state.
Download a fair-sized cache to avoid having to repeat this procedure too often. Completely shutdown BOINC. Using a plain text editor, open the file named client_state.xml in the data directory (check first few lines of messages tab in Manager for location). Do a global replace of <rsc_disk_bound>10000000 with <rsc_disk_bound>50000000
Save the file. Restart BOINC. Check the messages tab for error messages. If there aren't any you've successfully fixed the WUs. Otherwise, consider restoring from backup which I hope you made.
Raising the bound on unaffected WUs such as other LLRs or even WUs of other projects makes no difference.
Nice speedup indeed. Down to ~300 seconds from ~440 on my Q9550. :)
____________
 |
|
|
|
|
|
I got the error with clients i put on ppsLLR today:
number one
number two
|
|
|
