[Gems-users] L1 v.s. L2 cache miss in MESI_CMP_filter_directory protocol

[cge@xxxxxxxx]

Hi,
Does anyone know the relationship among  L2_cache_total_misses,
 L1D_cache_total_misses, and  L1I_cache_total_misses ?

I ran transactional microbenchmarks in MESI_CMP_filter_directory, and oberved
that L2_cache_total_misses is close to (L1D_cache_total_misses +   
L1I_cache_total_misses).
I suppose L2 cache miss should be a lot smaller than
L1D_cache_total_misses +  L1I_cache_total_misses.

Thanks,
Cen