CPU Hogging

Questions and Answers : Windows : CPU Hogging

Reply to this thread
Subscribe to this thread
Sort
AuthorMessage
Ron Parker
Forum moderator
Project administrator
Project developer
Avatar
Joined: Mar 27, 2007
Posts: 14
ID: 1
Credit: 1,232
RAC: 13
Message 2 - Posted 20 Jun 2007 7:24:21 UTC
Last modified: 20 Jun 2007 7:25:19 UTC

It has been noted that the BOINC client experiences high CPU usage when transferring files for SciLINC.

Augustine found a partial workaround to this. It is discussed on BOINCstats.

Julian Ellis
Joined: Jun 11, 2007
Posts: 2
ID: 65
Credit: 0
RAC: 0
Message 25 - Posted 21 Jun 2007 12:23:56 UTC

The reason for this is the number of files that are seperately downloaded/uploaded for each WU it is a massive waste of bandwidth resourses for both the project servers and the end users.

Wouldn't it be possible to put them into all into one file that can be transfered in 1 go (you could even consider compressing the file)

Ram Raider
Joined: Jun 12, 2007
Posts: 1
ID: 141
Credit: 0
RAC: 0
Message 28 - Posted 21 Jun 2007 18:52:18 UTC - in response to Message ID 25.

Actually the main reason for the hogging of the CPU is the size of the client_state xml file. BOINC updates this file on a regular basis and after every contact with a project, including file transfers and requests for work.

With about 150 workunits, the file size is approx 8MB and greater. On a slowish computer it takes about 20 seconds to update the file with the BOINC process running at 100% utilisation. As BPOINC runs at normal priority this makes the whole computer unresponsive as many people found out over the weekend.

I noticed this behaviour even when all the workunits had been downloaded and waiting to run becuase I'm attached to many projects that have no work. BOINC was barely able to keep up with polling for work round the projects!

Therefore the priority has to be to keep the size of the client_state xml file to a minimum which probably requires both zipping up the files and having longer workunits (so less are cached).



Ron Parker
Forum moderator
Project administrator
Project developer
Avatar
Joined: Mar 27, 2007
Posts: 14
ID: 1
Credit: 1,232
RAC: 13
Message 33 - Posted 21 Jun 2007 23:27:01 UTC - in response to Message ID 25.

Wouldn't it be possible to put them into all into one file that can be transfered in 1 go (you could even consider compressing the file)


Yes, we are working on that currently.

Ron Parker
Forum moderator
Project administrator
Project developer
Avatar
Joined: Mar 27, 2007
Posts: 14
ID: 1
Credit: 1,232
RAC: 13
Message 35 - Posted 21 Jun 2007 23:32:33 UTC - in response to Message ID 28.

Actually the main reason for the hogging of the CPU is the size of the client_state xml file. BOINC updates this file on a regular basis and after every contact with a project, including file transfers and requests for work.

With about 150 workunits, the file size is approx 8MB and greater. On a slowish computer it takes about 20 seconds to update the file with the BOINC process running at 100% utilisation.


Right and when we finish bundling all of the files for transfer this should reduce the size of client_state.xml by a factor of 50 or more. Currently the number of files sent per work unit is 50 due to a limitation in the size of some XML entries in the tables created by BOINC.

Once we zip the input and output files this 50 file limitation should go away.

vfrey
Joined: Jun 11, 2007
Posts: 1
ID: 59
Credit: 1
RAC: 0
Message 39 - Posted 22 Jun 2007 9:19:09 UTC

I'm not sure if this was already mentioned, but the big number of upload files is also a result of the very short calculation times of the WUs. As far as I recall each WU lasted only 1 second on my PC. I don't think data transfer makes sense at all with such short calculation times.

Reply to this thread

Questions and Answers : Windows : CPU Hogging



Return to SciLINC main page


Copyright © 2007 Missouri Botanical Garden
Written by Ron Parker