Message boards : Number crunching : New WUs writing too much to disk
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile yo2013
Avatar

Send message
Joined: 16 Oct 13
Posts: 59
Credit: 342,408
RAC: 0
Message 26927 - Posted: 28 Oct 2014, 11:03:41 UTC

Today's new WUs write to disk every 3-10 seconds, depending on the computer I use. Why so much writing?
ID: 26927 · Report as offensive     Reply Quote
genhos

Send message
Joined: 26 Sep 13
Posts: 3
Credit: 1,241,313
RAC: 0
Message 26928 - Posted: 30 Oct 2014, 8:56:16 UTC

I too am having issues with these w-b3_-26000 units where they crunch fine for around an hour, reach 100% then just continually write to the HDD. Using the Win8 task manager it shows 100% HDD write. Then the computer becomes immensely unresponsive. All of the unit's "Wingmen" have finished these fine but use Linux rather than the Win8.1 I'm using.
I have suspended these units for now and likely to have to abort them just to try and get some new work.

Any ideas?
ID: 26928 · Report as offensive     Reply Quote
waveybarrel

Send message
Joined: 15 Oct 13
Posts: 6
Credit: 28,625
RAC: 0
Message 26929 - Posted: 30 Oct 2014, 11:04:53 UTC
Last modified: 30 Oct 2014, 11:05:46 UTC

Also seem to use a lot of disk space until it completes. This task http://lhcathomeclassic.cern.ch/sixtrack/result.php?resultid=48058882 was up to 600MB just before completing.
ID: 26929 · Report as offensive     Reply Quote
genhos

Send message
Joined: 26 Sep 13
Posts: 3
Credit: 1,241,313
RAC: 0
Message 26930 - Posted: 31 Oct 2014, 0:12:35 UTC

I have been able to leave these units running while away from the computer for a few hours and it some have now been returned and validated. Shame about the large amount of disk write that meant that I couldn't do these units without significantly affecting the computer use during the normal day.
ID: 26930 · Report as offensive     Reply Quote
Eric Mcintosh
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 12 Jul 11
Posts: 857
Credit: 1,619,050
RAC: 0
Message 26931 - Posted: 31 Oct 2014, 13:18:25 UTC

Sorry about this. These WUs require a very precise
estimation of the DA which implies writing the particle
co-ordinates every turn to 30 files (one file per pair
of particles), about 100 bytes each time. This data is then used by
the post-processing to make the results summary. I also
endfile and backspace after each write to flush the buffer.
This seems to be OK on Linux but from your comments appears
to be a problem on Windows. The total data is indeed about
600MB for 100,000 turns, 30 particle pairs.

I am wondering if I can remove the endfile/backspace except before
doing a checkpoint. Difficult to change right now as we have a lot
of work to process before he upcoming HL-LHC conference.

I shall see what I can do, but I shall be on vacation for two
weeks and will have no remote access to my Windows box for
testing and I am reluctant to modify a well proven procedure. Eric.
ID: 26931 · Report as offensive     Reply Quote
genhos

Send message
Joined: 26 Sep 13
Posts: 3
Credit: 1,241,313
RAC: 0
Message 26935 - Posted: 31 Oct 2014, 14:46:28 UTC

Thanks for the reply Eric.
All of mine have now cleared through my system. During the actual crunching, the write didn't seem to be a problem it was when it had completed the unit that the writing made the computer almost unusable. Leaving the machine alone for a few hours allowed the write to complete and then upload the result, unfortunately I don't know how long this took as I was away from the computer.
The computer used for this batch of units was a Win8.1 Pro 32bit. Can't say I have seen this happen on any previous units.
Each of the units I had, CPU time was around 3,500 seconds but run time varied from 4,000 to 6,000 seconds. WU numbers 22544559 - 22544565 and 22544603 -22544610.
ID: 26935 · Report as offensive     Reply Quote
Eric Mcintosh
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 12 Jul 11
Posts: 857
Credit: 1,619,050
RAC: 0
Message 26936 - Posted: 31 Oct 2014, 15:31:07 UTC - in response to Message 26935.  

Thanks for the feedback; I am running 4 such WUs on my Windows 7
with 4 threads BUT on Cygwin.......seems OK but it then
takes a VERY long time to finish. Investigation proceeds.
Eric.
ID: 26936 · Report as offensive     Reply Quote
Eric Mcintosh
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 12 Jul 11
Posts: 857
Credit: 1,619,050
RAC: 0
Message 26937 - Posted: 31 Oct 2014, 16:23:12 UTC

First look seems to eliminate the "tracking".
My suspicions are now centered on the post-processing
phase where 30 files will be read. Hopefully this
will take at most some minutes. Investigation
continues; can't make changes late on Friday!
ID: 26937 · Report as offensive     Reply Quote
Profile Tom95134

Send message
Joined: 4 May 07
Posts: 250
Credit: 826,541
RAC: 0
Message 26938 - Posted: 31 Oct 2014, 16:58:58 UTC

At one time I did "real work" (business) on my Windowz system. Mainly using Photoshop, Illustrator, and Framemaker. I put these applications in the BOINC->Preferences->Exclusive applications tab. This seemed to work very well for me. Now I don't do a lot of "heavy lifting" so the slowdown is something I can live with.

The only thing I've noticed is the Exclusive applications option may not suspend vLHC crunching. (But I haven't really looked closely.)
ID: 26938 · Report as offensive     Reply Quote

Message boards : Number crunching : New WUs writing too much to disk


©2024 CERN