The people you have to test this are not your target audience,  The 
testars are self-selected volunteer computing zealots who would rather 
have broken video playback (or stop BOINC manually) than a 5% drop in RAC.

On 2/17/2010 10:05 AM, David Anderson wrote:
>
> Rom Walton wrote:
>>
>> It seems to me that it would be better to monitor this once a second and
>> then use a decaying average to prevent needlessly starting and stopping
>> processes for apps that jump around the user defined threshold.
>
> I'm not sure what you mean by "needless".
> If the load average is close to threshold,
> BOINC should stop until it's well below threshold.
>
> The current policy is:
> every 10 seconds, look at CPU usage over the last 10 sec.
> If it's greater than 25%, suspend BOINC; otherwise, resume BOINC.
>
> Suppose there's some activity that uses 100% of the CPU
> for 1 second, every 5 seconds.
> The current mechanism won't trigger.
>
> a) We could make it trigger by sampling every 1 sec.
> Then, on average, BOINC would suspend itself halfway
> through every spike.
>
> b) Or we could be more aggressive: sample every 1 sec,
> and if CPU load is above 25%, suspend BOINC for the next 10 sec.
> This would keep BOINC suspended indefinitely while
> that activity is going on.
>
> We need to do some experimentation with real apps
> (e.g. video playback, commercial-removing software)
> to decide whether to use a) or b), and what the parameters should be.
> Maybe what we should do is provide detailed controls via cc_config.xml,
> and let people experiment.
>
> -- David
> _______________________________________________
> 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.
>
_______________________________________________
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