Dear All, I've built a cluster to test the IO500 benchmark. The preliminary results showed that the new "shift" parameter did influence the stat result badly. If I use "-N" parameter, the mdtest result is: mdtest_easy_stat 1429.930 kiops : time 253.86 seconds mdtest_hard_stat 962.500 kiops : time 191.10 seconds If I don't use "-N" parameter, the mdtest result is: mdtest_easy_stat 200250.000 kiops : time 1.84 seconds mdtest_hard_stat 51038.200 kiops : time 4.47 seconds It is clear that the "shift" parameter change will impact the stat result with the same hardware and software configuration. Now I'm still trying to optimize the test. Hope I'll deliver a better result later. Ruibo Wang