It's not that there's a different amount of "work." The execution
paths (and therefore the number of cycles, number of instructions,
number of memory references, number of messages, etc.) might be
different. The work should be the same, assuming that we're defining
work as useful computation from the application's point of view.
...Greg
Lide Duan wrote:
I implemented dimensional order routing, some kind of
minimal adaptive routing and also a fully adaptive routing algorithms
for torus networks. But according to Mike, the different amount of msgs
may be due to the different amount of work between the two magic
instructions when the different routing algorithms are utilized...
On 11/1/07, Niket <niketa@xxxxxxxxxxxxx> wrote:
Can
you tell more about the changes that you made.
I think that the interconnect might generate different amount of
messages due to routing.
Also, I am surprised that running Ruby twice gave you exactly the same
result. It is surprising that the OS did not cause any randomizations.
-Niket
Lide Duan wrote:
> Hello there,
>
> I have got some confusion about Ruby running on the same
checkpoint...
>
> Basically I modified the routing algorithm utilized in the on-chip
> interconnect of Ruby, and compared the output results with the
> unmodified Ruby version. For both simulations, the checkpoint has
been
> run from one magic instruction to the other, implying the same
amount
> of work. I assumed that the msgs injected into the interconnect
(from
> the components) should be the same for both cases because the only
> difference is the routing algorithm, which should not affect the
> behaviors on the cache level since all the msgs finally arrive to
> their corresponding destinations.
>
> However, the numbers of the injected msgs of the two runs have 5%
> difference, and the msgs generated in the interconnect (all of them
> are Invalidate_Control, to my observation) have more the 10%
> difference. The numbers of instructions of the two simulations are
> also different (but not much) although the two runs were from the
same
> start point to the same end point.
>
> On the other hand, if I run Ruby twice without making any
modification
> between the two runs (
i.e. the same Ruby on the same workload), the
> results are exactly the same. So I don't think there is any random
> effect to cause the above difference. Then, are the results I have
got
> reasonable?
>
> Thanks,
> Lide
>
------------------------------------------------------------------------
>
> _______________________________________________
> 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.
>
>
_______________________________________________
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.
_______________________________________________
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.
|
|