[Gems-users] MLP-Memory level parallelism

[Muhammad abid Mughal <mabidm_pieas@xxxxxxxxx>]

Hi guys,

            Hope you all doing good I ran couple of simulations with Opal+Ruby to measure MLP. But in each case Avg. MLP was so high (multiple of hundred and thousand). I am wondering the way Opal calculate  MLP (in rubycache_t class)  is not correct. Here is the description how Opal calculates MLP :

        for each miss it calculates the difference of cycles = current miss time - m_last miss time

                                                     localsum =  difference of cycles * m_tot_miss_outstanding

                                                     m_totalmlpsum += localsum

       variables start with m_ are data members of rubycache_t class

     Finally it divides  m_totalmlpsum/total_local_cycles_core 

     to calculate Avg. MLP.

Any comments?

Muhammad abid