There would also need to be some history tracking as well ... in some cases the prior tasks run are what set up the issues in the first place ...
It might be sufficient to track only the last hour, though I would suggest that this be a configurable option ... One other thing, looking at the documentation of the simulator (http://bossa.berkeley.edu/trac/wiki/ClientSim) many of the terms are not defined well... There also appears to be an implicit assumption that the Number of GPUs will not be ATI: >0 and Nvidia <> 0; or, to be more specific to be able to model the system have in W02 where I have one GTX260 and an HD5870 ... I note this because the GPU RAM is a single number as is the FLOPs est ... in the system I just described the cards have 875 and 1024 RAM respectfully and different FLOPs capabilities by a factor of about 3-5 (depending on the project) So, were I to try to model DNETC's actions on this machine I would not be able to model it correctly... because the same tasks have a world of difference between the way they run on CUDA and on the ATI cards ... also, how do you inform the sim that DNETC runs on all like GPUs at the same time in contrast to other projects (essentially MP on alike GPUs) Hmm, looking at the instruction capabilities based on the input files you don't actually model real world work loads but model based on assumed inputs with what are essentially fixed length tasks with some slight vaiability ... this is inadequate for a large number of projects, for example RCN tasks take seconds to a full day to complete as do ABC and quite a few other projects ... it would be far more instructive to be able to model based on actual work performed.... In this case, the generate_sim_model flag would be set, the projects and hosts files would be generated ... then a log would be made of the actual tasks obtained and run ... this would allow you to actually see what happened ... Last point, the sim policy is 5.10.13? there have been a world of changes since then ... On May 5, 2010, at 8:36 AM, [email protected] wrote: > > It might be useful for testing problem CPU scheduling cases to have a > "Generate Scenario XML" button someplace that would generate an XML file > suitable for input into the simulator. This owuld entail tracking some > extra information about the projects so that the schedule would know about > the projects attached. > > 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.
