Making rr_sim do a simulation of exactly what will happen is going to be very difficult as there are multiple sets of tasks that can be run at each task switch opportunity based on which tasks are tagged as needing high priority by running rr_sim at the time of the task switch opportunity. This would tand to make rr_sim recursive and take O(n!) time in the worst case. Work fetch just needs to take into account the worst possible arrangement of tasks into CPUs over time and assume that is what is going to happen and make certain that there is enough work on hand based on this worst case to get through the disconnected interval.
I believe that the worst possible arrangement is least remaining runtime first as this pushes the tasks with the most run time to occupy a CPU as late as possible. This can be most easily done as I have described before. 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.
