Re: [Gems-users] Some questions about the inter-connect network

[Lei Jin <jinlei@xxxxxxxxxxx>]

Mike,

Thank you for your quick response.

Based on my understanding of your explanation of "chip", if I want to 
make a CMP with a single shared L2, then I should just make one chip? Is 
this right?
In the 2DTorus topology, the makeSwitchesPerChip() just connects all the 
MessageBuffers to the same switch for that chip. Then you connect the 
switches. I feel the 2D Torus topology is really among chips. But from 
your explanation, the topology is implemented among processors. Can you 
explain this a little bit? Especially how do you calculate the number of 
switches by following line:
numberOfTorusSwitches = m_nodes/MachineType_base_level(MachineType_NUM);

Thanks!

Lei

Mike Marty wrote:
> > I am a little confused by the inter-connect networks implemented in
> > GEMS. What does the "chip" stand for? Is it a CMP? Or just a core within
> > the CMP?
> A Chip in Ruby is really just a way to organize some parameters and to
> keep track of instances of objects.  We usually think of it as number of
> processors within a CMP, but you don't have to.
>
> For example, it is used by the CMP protocols to denote which processors
> belong to the shared-L2 cluster.
>
> however if your target CMP has multiple shared L2s on the same CMP, then
> you could still model this as one CMP even though Ruby has several "Chip"
> instantiations".
>
> > I guess the organization here is in a computer system, we can have
> > multiple CMP "chips", then within each chip, we have multiple cores. Is
> > it right?
> >
> > If it is correct, I don't understand why you use
> > m_nodes/MachineType_base_level(MachineType_NUM) to get the
> > numberOfTorusSwitches. And you say it is number of inter-chip switches.
> >   How do you connect the components with chip?
> The interconnect model doesn't care about Chips.  It just connects
> together MessageBuffers which attach to cache controllers.
>
> Yes, its confusing and is an artifact of refactoring our old SMP-only code
> to remain backwards compatible with our older SMP protocols (that were not
> released).
>
> _______________________________________________
> 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.