Braulio,
In multiprocessor systems, the execution path (and the number of
instructions) will vary from execution to execution, depending on (among
other things) minor variations in the observed memory latencies. Even
small deltas can cause locks to be acquired in different orders, causing
different executions. These locks can reside in the OS, remember, so
even single-threaded code can suffer on multithreaded machines. We
combat this variation by running multiple simulations of each
configuration and computing 95% confidence intervals.
As far as comparing the number of instructions executed, you might want
to ensure that simulation really has stopped in the same "place" -- eg.
the insturction pointers of each processor are identical for both of
your protocols when execution finishes. When using magic instructions in
more than one location, problems can arise when they are reached in
different orders.
Regards,
Dan
Braulio Pino wrote:
Hello,
I have run two different implementations of the MOSI protocol and I want to
check whether one of them improves the other in terms of execution cycles.
To this end, I have created a checkpoint and I use the magic instructions in
order to the two simulations run the same part of the application that I
want to simulate. After running them, I have seen that the number of
executed instructions is different in spite of simulating the same part of
the application. Therefore, I do not know if it is possible that I have
done something wrong or, on the other hand, those results could be right. I
think that if I simulate the same part of the code, both simulations should
run the same number of instructions, shouldn't they? In case the simulations
are correct, do I have to take into account the number of executed
instructions to compare the execution time or I can compare the ruby cycles
directly?
Thanks you.
Regards.
_________________________________________________________________
Horóscopo, tarot, numerología... Escucha lo que te dicen los astros.
http://astrocentro.msn.es/
_______________________________________________
Gems-users mailing list
Gems-users@xxxxxxxxxxx
https://lists.cs.wisc.edu/mailman/listinfo/gems-users
Use Google to search the GEMS Users mailing list by adding "site:https://lists.cs.wisc.edu/archive/gems-users/" to your search.
|