What type of benchmarks/applications are you running? From our testing, the major reason for high variation in MARSS is due to variation in benchmark's thread to core scheduling between runs. If you are running multithreaded applications then you should set affinities of each thread.
- Avadh On Mon, Sep 24, 2012 at 5:34 PM, ali ahmadi <[email protected]>wrote: > Hi All, > > > I ran some codes using MARSS and every time I ran (configuration and input > data are same) cache states are different. I used ooo core with 16KB cache > for data and 16 Kb for instruction. > > Here are states in three time run of a program with same data and same > configuration: > > > > i cache hit: 37402422 33482838 35464745 > i cache miss 202043 710565 1020877 > > > LD > D cache hit: 74044556 73879784 73605248 > D cache miss 158954 285485 266320 > > > ST > D cache hit: 31369883 31991901 31815062 > D cache miss 97498 130096 99196 > > > Program simulated for more than 400,000,000 cycles. > > I really appreciate your comments and help! > > > > Regards, > Ali > > _______________________________________________ > http://www.marss86.org > Marss86-Devel mailing list > [email protected] > https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel > >
_______________________________________________ http://www.marss86.org Marss86-Devel mailing list [email protected] https://www.cs.binghamton.edu/mailman/listinfo/marss86-devel
