Re: [Gems-users] Detailed Network in GEMS


Date: Wed, 21 Mar 2007 13:28:25 -0400
From: Niket Agarwal <niketa@xxxxxxxxxxxxx>
Subject: Re: [Gems-users] Detailed Network in GEMS
Hi Mike. Thanks for the reply. Guessing by your answer, what I gather is that, if I interface my detailed Network Simulator, I will have to ensure that the memory transaction will complete through my network simulation and not through the in-built network. What I want to point out is that inserting my detailed network will effect the finishing time of a memory transaction and hence the simulation.
Am I right ?

Niket.





____________________________________________________________________________
Niket,

Yes, I believe that interfacing your own network simulator into
PerfectSwitch::wakeup() is a sound strategy.  Thus the endpoints of the
interconnect has PerfectSwitches, and everything in between is your own
simulation.

I imagine your network simulator will likely just deliver messages to
their endpoints at the right cycle.

GEMS ultimately unstalls Simics when a memory transaction completes.  It
doesn't really return a certain stall time to Simics.

--Mike


Hello all,
I am trying to write my own detailed network simulation inside GEMS. The
way I have planned is:
There is a function in PerfectSwitch::wakeup(). This is called whenever
any message is enqueued into the switch's buffer. The thing that GEMS
requires is the stall time. What I plan is to trap these messages from
this point and interface them into my detailed network simulator, which
in turn will return the stall time. Are there issues in this ?

Mike might throw some light into this.

Thanks,
Niket.
____________________________________________

[← Prev in Thread] Current Thread [Next in Thread→]