If min_rpc_delay < last_rpc_duration*5, then use a backoff for the project up to something like max(100*last_rpc_duration, 100*min_rpc_delay). In the event that last_rpc_duration*5 > min_rpc_delay, clear the backoff.
This will force projects that use a min_rpc_delay setting is appropriate for their infrastructure and will automatically protect other projects. It will also help projects that have a temporary overload. > >From: David Anderson <[email protected]> > > > >Can anyone think of a solution for this? > >-- David > > > >On 11-Aug-2012 11:40 AM, [email protected] wrote: > >> OK, how about this scenario for a problem. > >> > >> low latency project has a current value of 1 minute for min_rpc_delay.? It > >> takes 10 seconds for an RPC. > >> SETI@Home is overloaded and takes 90 seconds to respond to an RPD delay. > >> Neither is currently handing out work. > >> > >> Result:? Any host connected to both where both are the top two for work > >> fetch is going to have some difficulty fetching work from anywhere else. > >> > >> jm7 > >> > >> _______________________________________________ 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.
