I am a naive user of TBB with my intel centrino duo machine.
I am currently doing a project of parallelizing Sudoku solving.
I have accomplished the program with my 2 cores.
I have the following questions on benchmarking its performance:
1)Can we decide the number of cores that TBB can operate on? This is to see the speed up happening between using 1 core and 2 cores. By default it is using both the cores.
2)I have only 2 cores. But i would also like to see or estimate how much will be the time taken when operated on more cores. Is there any emulators for multicore or any formula by which i can estimate the time taken when more cores are added? Or how else can i do the same?
3)Can i calculate the runtime memory usage by my algorithm using TBB?
1. Yes, you can pass desired number of worker threads to constructor of tbb::task_scheduler_init.
2. Well, there is Intel Parallel Advisor which can kind of predict scalability for higher number of cores. However I suggest to run the program on a real machine, that will give more precise results. You may use Intel Parallel Universe service to test your program on a machine with 16 hardware threads: http://paralleluniverse.intel.com/