Dear Niko,

Thank you for your answer! What I meant with cache partitioning is the
dynamic partitioning of the ways of a shared LLC between the processors of a
CMP system using various policies/metrics. Sorry for the inconvenience.
However, I have found the modules that I should modify in order to implement
this algorithm, so that's ok.

One question on the "make" of the CMPFlex simulator: When I try to compile
CMPFlex I get the message:

/usr/include/c++/4.2/x86_64-linux-gnu/bits/c++config.h:149: error: expected
constructor, destructor, or type conversion before 'namespace'

Any clue on what might be the problem? I have used the gcc and simics
versions proposed in the Simflex tutorial..

Thank you for your help,
Christina


2009/4/15 Nikos Hardavellas <hardav at cs.cmu.edu>

>  Dear Christina,
>
>
>
> I assume that by ?partitioning? you mean dividing the cache into smaller
> pieces and assigning each piece to a core (e.g., a NUCA cache).
>
> In the current SimFlex release, the cache is not partitioned among the
> cores; rather, it looks like a single structure that all cores share.
>
> To create a NUCA cache, you need to instantiate multiple objects of the
> cache (each object being a separate cache partition), connect them together
> through an on-chip interconnection network using NetShim, and assign each
> cache partition to a core. Care must be taken to maintain data coherence
> among the partitions, avoid deadlock (i.e., have separate channels for
> requests and replies to break dependency cycles) and do the routing properly
> (e.g., configure NetShim with a correct network topology and do
> dimensionality minimal source routing to avoid deadlocks at the network).
>
>
>
> If by ?partitioning? you mean something different, please give us more
> details.
>
>
>
> Regards,
>
> --nikos
>
>
>
>
>
> *Nikos Hardavellas*
> Carnegie Mellon - Computer Science Department
> 5000 Forbes Ave, Pittsburgh, PA 15213, USA
> +1 (412) 268-5005   ECE - Hamerschlag Hall A312
> http://www.cs.cmu.edu/~hardav <http://www.cs.cmu.edu/%7Ehardav>*
> **hardavellas at cmu.edu*
>
> *From:* simflex-bounces at ece.cmu.edu [mailto:simflex-bounces at 
> ece.cmu.edu] *On
> Behalf Of *Christina Delimitrou
> *Sent:* Wednesday, April 08, 2009 7:36 AM
> *To:* simflex at ece.cmu.edu
> *Subject:* [Simflex] Cache partitioning policies using Simflex
>
>
>
> Hello,
>
> I am currently working on cache partitioning policies for CMP architectures
> and wanted to ask if there is a way to use Simflex to achieve a faster
> simulation for these schemes. I imagine that I will need to modify the
> simulators' modules CMPFlex and TraceFlex to partition the shared cache
> among the cores and I was wondering if I also need to change the
> implementation of the components FastCache or CmpCache.. (1) Is there any
> prior implementation for a cache partitioning algorithm in Simflex and (2)
> was is the default cache partitioning scheme you use?
>
> Thank you,
> Christina
>
> --
> Undergraduate Student
> Electrical and Computer Engineering Department
> National Technical University of Athens
> Computer Systems Lab
> +30 210 6513839, +30 6947424452
> http://users.ntua.gr/el04019/Homepage/Home.html
>
> _______________________________________________
> SimFlex mailing list
> SimFlex at ece.cmu.edu
> https://sos.ece.cmu.edu/mailman/listinfo/simflex
> SimFlex web page: 
> http://www.ece.cmu.edu/~simflex<http://www.ece.cmu.edu/%7Esimflex>
>
>


-- 
Undergraduate Student
Electrical and Computer Engineering Department
National Technical University of Athens
Computer Systems Lab
+30 210 6513839, +30 6947424452
http://users.ntua.gr/el04019/Homepage/Home.html
-------------- next part --------------
An HTML attachment was scrubbed...
URL: 
<http://sos.ece.cmu.edu/pipermail/simflex/attachments/20090511/4c8776af/attachment.html>

Reply via email to