James,
I've wondered that myself. One place to start would be to
vary the network topology so that memory is always the
number of networks hops. You can do that with the file
specified network though I'm not an expert on that works.
Or, you could simply change the network latencies for each
system image so that the larger machines, which incur more
hops, run each hop faster. But, I suspect there will still
be other overheads, such as more OS activity, when you run
larger target machines.
--Kevin
James Wang wrote:
Hi,
I was wondering if anyone has any ideas how to make an efficient and
accurate simulation of different numbers of threads on Ruby...
What I mean is, I would like to have the results of 1 thread running
on 1 processor be similar to the results that I would obtain from
having 1 thread running on a 32 processors machine, but only using one
of the 32 (binding it).
The obvious solution would be to always use a 32 processor machine,
regardless of the number of threads. But that, as you might know, will
result in a significantly slower simulation. Running 1 thread on 1
processor, 2 threads and 2 processor, without taking the scalability
of our programs into consideration skews the results in favour of the
machines with a lower number of processors due to (as I understand)
the way the interconnection network is simulated.
Is there an easy way of somehow adjusting the 1 processor image to
perform the same as it would assuming it had 32? Or a way to always
use 32 processors while somehow speeding up the simulation to ignore
the ones that aren't being used?
Thanks in advance for any reply. :)
Regards
James
_______________________________________________
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.
|