Hi Mike,
I tried printing the topology by setting the
g_PRINT_TOPOLOGY to true. But then the output file
prints the following line.
"Topology print ONLY indicates the _NETWORK_ latency
between two machines
It does NOT include the latency within the machines"
In our case we are simulating a 16 core/single chip
CMP. Is the printed intermachine latencies are still
applicable for our model?
Besided what is the abstract name used to represent
onchip-directory? In our printed file we see only
L1cache<no.>, L2Cache<bank no.> and Directory
(offchip).
Thank you
--n
--- Mike Marty <mikem@xxxxxxxxxxx> wrote:
> The network topology is independent of the messages
> a protocol uses. The
> PT_TO_PT and FILE_SPECIFIED are appropriate for this
> CMP protocol.
>
> To see the _network_ latencies, set g_PRINT_TOPOLOGY
> to true in
> $GEMS/ruby/config/rubyconfig.defaults. Then in the
> Ruby stats, you will
> see lines like the following to give you the minimum
> latency between
> components:
>
> L1Cache-0 -> L2Cache-0 net_lat: 5
>
> PT_TO_PT creates a simple network and uses the
> NETWORK_LINK_LATENCY and
> ON_CHIP_LINK_LATENCY.
>
> FILE_SPECIFIED networks use the files in
> $GEMS/ruby/network/simple/NetworkFiles. You can
> create an arbitrary
> topology with link latencies and bandwidths
> specified. The set of network
> files is chosen based on the g_CACHE_DESIGN
> parameter.
>
> --Mike
>
>
> > Hi Mike,
> > Thanks for the immediate response. How to
> calculate
> > the network latency between an L1 cache to the
> onchip
> > directory? Will these transfers (say an unblock
> > message) happen on the regular network defined by
> > Topology?
> >
> > thank u
> > --n
> >
> > --- Mike Marty <mikem@xxxxxxxxxxx> wrote:
> >
> > > MOESI_CMP_directory assumes shadowed L1 tags for
> > > protocol logic, but
> > > completely abstracts away the implementation
> details
> > > of keeping a shadowed
> > > copy by using a "perfect directory" structure
> for
> > > implementation in SLICC.
> > >
> > >
> > > > hi,
> > > > I am using MOESI_CMP_directory protocol and
> I
> > > would
> > > > like to know whether it employs distributed
> (one
> > > per
> > > > L2 bank) or centralized on chip directory. If
> it
> > > is
> > > > distributed how L1 tags are tracked?
> > > >
> > > > --n
> > > >
> > > >
> > > >
> > > >
> > > > __________________________________
> > > > Yahoo! Mail - PC Magazine Editors' Choice 2005
> > > > http://mail.yahoo.com
> > > >
> _______________________________________________
> > > > Gems-users mailing list
> > > > Gems-users@xxxxxxxxxxx
> > > >
> > >
> >
>
https://lists.cs.wisc.edu/mailman/listinfo/gems-users
> > > >
> > > _______________________________________________
> > > Gems-users mailing list
> > > Gems-users@xxxxxxxxxxx
> > >
> >
>
https://lists.cs.wisc.edu/mailman/listinfo/gems-users
> > >
> >
> >
> >
> >
> >
> > __________________________________
> > Yahoo! Mail - PC Magazine Editors' Choice 2005
> > http://mail.yahoo.com
> > _______________________________________________
> > Gems-users mailing list
> > Gems-users@xxxxxxxxxxx
> >
>
https://lists.cs.wisc.edu/mailman/listinfo/gems-users
> >
> _______________________________________________
> Gems-users mailing list
> Gems-users@xxxxxxxxxxx
>
https://lists.cs.wisc.edu/mailman/listinfo/gems-users
>
__________________________________
Yahoo! Mail - PC Magazine Editors' Choice 2005
http://mail.yahoo.com
|