The BOINC client alternates between CPU and GPU requests when communicating
with a project.  But, if the project doesn't have a GPU application, there
is a lot of bandwidth wasted by clients making GPU requests.  If there isn't
already a better way to handle this, could the client cache the fact that
project X doesn't have a GPU app and only request CPU for the next X
hours/days?    

 

For example, once per day (or some time frame set by the client in the
preferences), the boinc client could get the data from
<project_url>/apps.php?xml=1  and use that to determine which types of apps
are available by looking at the plan classes.  It would ignore the dates,
versions, etc.  since those would change more often.  If there aren't any
apps with CUDA, ATI, or OPENCL plan classes, then there's no need to ask for
GPU work.  If a project uses relatively small workunits and has "resend lost
workunits" enabled, there can be a large amount of network traffic wasted.
Or am I missing something with the way the scheduler works?

 

Jon Sonntag

_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.

Reply via email to