I checked in code that keeps track of the fraction of time when both the CPU and network are available, and uses this to estimate the rate of progress of jobs with the needs_network flag.
The CERN T4T app is unusual in that runs for 24 hours and exits, regardless of how much computing was done. BOINC's mechanisms for estimating runtime are a bit pointless since the runtime is predetermined. If we wanted to handle the T4T app optimally, we'd need a mechanism for telling the BOINC client about the 24 hour runtime; this doesn't currently exist. -- David On 26-Jun-2012 9:39 AM, [email protected] wrote: > No, the client is not running the task, however, I think I know what is > going on. > > VBoxWrapper only gets time when BOINC is allowed to use the internet. This > machine has to have BOINC disallowed from using the internet for 14 hours > per day. > > BOINC assumes that all projects will be allowed to process the same > fraction of the time (up_time * on_time * efficiency). There are now at > least three distinct groups of applications that have different "on_time" > values. CPU based applications, for which the current math works just > fine. GPU based apps, which may need a different value for on_time, and > applications that require the network. > > The three are distinct based on different settings available for > controlling when they run. There are separate settings that allow CPU and > GPU applications to run if the user is active. On many machines, having a > CPU application run in the background works well, while on the same machine > allowing a GPU application to run in the background almost freezes the > machine for the active user. Applications that require the network can, of > course, only run while the network is actually allowed. There may be a > fourth set that is GPU and network, but I do not believe that there are any > applications in this set yet. It is also true that some applications and > projects are much more efficient at using runtime than others. > > That said, I would propose that rr_sim be changed so that each application > has its own calculation of the fraction of time that it gets to run based > on which category it falls into (CPU / GPU, needs network / no network, and > whatever other factors arise in the future). > > It might also be possible that the applications that need network access > only need it occasionally. If this is the case, then it might be a good > idea for the application to be able to change its state on the fly. If it > is in the state where it does not need network access when the network is > disabled, it could run until such a time as it actually needs network > access to continue - at which point it would be swapped out. When the > network was enabled, it would be rescheduled as normal... > > jm7 > > > |------------> > | From: | > |------------> > > >--------------------------------------------------------------------------------------------------------------------------------------------------| > |David Anderson<[email protected]> > | > > >--------------------------------------------------------------------------------------------------------------------------------------------------| > |------------> > | To: | > |------------> > > >--------------------------------------------------------------------------------------------------------------------------------------------------| > |<[email protected]> > | > > >--------------------------------------------------------------------------------------------------------------------------------------------------| > |------------> > | Cc: | > |------------> > > >--------------------------------------------------------------------------------------------------------------------------------------------------| > |<[email protected]> > | > > >--------------------------------------------------------------------------------------------------------------------------------------------------| > |------------> > | Date: | > |------------> > > >--------------------------------------------------------------------------------------------------------------------------------------------------| > |06/22/2012 03:56 PM > | > > >--------------------------------------------------------------------------------------------------------------------------------------------------| > |------------> > | Subject: | > |------------> > > >--------------------------------------------------------------------------------------------------------------------------------------------------| > |Re: [boinc_alpha] 7.0.25 not starting a task that is overdue. > | > > >--------------------------------------------------------------------------------------------------------------------------------------------------| > |------------> > | Sent by: | > |------------> > > >--------------------------------------------------------------------------------------------------------------------------------------------------| > |<[email protected]> > | > > >--------------------------------------------------------------------------------------------------------------------------------------------------| > > > > > > > > On 22-Jun-2012 10:40 AM, [email protected] wrote: >> >> The task for CernVBox Wrapper is already past due, and is not getting any >> CPU time. > > Is the client running the task? > That's as much as BOINC can do. > The Cern app uses its own internal mechanism for > fetching and running jobs, > and if it's running but not getting CPU time, > then the problem lies there. > > -- David > _______________________________________________ > boinc_alpha mailing list > [email protected] > http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_alpha > 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.
