A financial risk management program is sped up by making 85% of the application concurrent, while 15% remains sequential. However, it turns out that during a concurrent execution the number of cache misses grows in a way dependent on N, the number of cores used. The dependency is CacheMiss =  Profiling the program reveals that 20% of the operations performed are memory accesses for both the sequential and parallel parts. The cost of other operations, including cache accesses, is 1 unit, and accessing memory has a cost of 3N + 11 units for the parallel part and a cost of 14 for the sequential part. Compute the optimal number of processors on which the program should run.

 

N N+10-