[Gems-users] Directories in MOESI_CMP_NUCA

["Mishali Naik" <mishalinaik@xxxxxxxxx>]

Hi All,
            I have a few questions regarding the MOESI_CMP_NUCA
protocol, I have looked at the generated code in L1Cache_Controller,
L2Cache_Controller and but I am still not clear:

1) In ruby/network/simple/Network_Files - the topology files have # of
directories equal to the # of processors. But the latency to each of
the directories is the same, so how are these 8 directories (in case
of 8 processors) used?

2) Does directory maintain coherence for L1? or L2? both? or none?

3) In case of DNUCA: What happens on a L2 miss? Is the request first
sent to the directory?

4) In case of SNUCA: how is directory used since the blocks are
statically placed?
        I appreciate all your help.

thanks,
Mishali
CS Undergraduate, UCLA