"Last month's column looked at bare context switch times by using the best primitives on both Windows and Linux. According to those results, context switch time under Windows takes only half as long as under Linux. I concluded that the algorithm for choosing a thread to run is well optimized in Windows.
"This month's column follows up on two issues. In the first, we look at giving the scheduler algorithm multiple choices by inserting more than one extra lock into the scheduling stew. Here again, we look only at the switching algorithm and give the processor no real work to accomplish. In the second issue, we finally look at accomplishing real work with threads. My workload consists of an intensive CPU computation of a fractal point. I give each thread a fractal point to compute and see how many iterations each can accomplish after a given amount of time. Will lots of threads get equal amounts of time?
"Recall last month's column showed that using the proper process synchronization primitives can make a major difference in the amount of time required to perform a context switch. Choosing the best primitives on each platform to measure the context switch time produces a reasonably fair comparison. Although last month's program accomplished no work, it did demonstrate the speed of context switching in both systems..."
