hi all,
I have coded up a new coherence protocol and I am running simulations
to evaluate it. I consider ruby cycles as the best variable/metric to
evaluate the overall speedup that the whole system will have. My
questions are:
1) I run the same simulation many times and I always get the same
number of cycles -- isn't Gems supposed to be non-deterministic?
(although ruby cycles are always the same, other variables are not)
2) when trying to compare the baseline ruby cycles VS my protocol's
ruby cycles, the comparison doesn't make sense. That's because the
number of instructions executed (instruction_executed) is very
different in these 2 simulations. Also, given that no matter how many
times i run a simulation i always get the same results, i feel like
comparing apples and oranges.. How can i approach this problem?
thanks a bunch
-Kostas
|