Hello, I confused by the results that the performance of Fib.x10 becomes worse while more cores are used to work. How come?
I try to use Fib.x10 to demonstrate the scalability of X10 on multi-core PC. I run it on a standalone multi-core 64-bit Linux box. And use both Java and C++ back-end. I use the -INIT_THREADS and X10_NTHREADS environment variable to control the thread pool size, which in turn can control how many cores are used. I expect that as more cores are used, the performance should become better. This is true with Java back-end( IBM J9). However, with C++, the result is total opposite. The performance become worse while the number of used cores increases. I compiled the X10 with -Doptimize=true and -DNO_CHECKS=true options. And compile the Fib.x10 with -O -NO_CHECKS options. I run the application using "mpirun -np 1 <others>". Other interesting result is that Java back-end have better performance than C++ back-end. Why more cores, worse performance for C++ back-end on a simple Fib? How C++ back-end is slower than Java back-end? I read the documents in X10 Day but still cannot figure out what's wrong. So Hope you can help me out. Thanks a lot. Regards, -Tetsu ------------------------------------------------------------------------------ This SF.net email is sponsored by Sprint What will you do first with EVO, the first 4G phone? Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first _______________________________________________ X10-users mailing list X10-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/x10-users
