Re: [Gems-users] Regarding ruby load-caches

[Polina Dudnik <pdudnik@xxxxxxxxx>]

Sparsh,

Instead of replaying the coherence requests to warm up the cache, what you could do is dump the contents of the cache into a file when you feel that it is sufficiently warm. So, the next time instead of warming it up by replaying the coherence requests, you can just load the cache with these values. 

Note however, that the configuration must match, so if your dumped cache contents are for 4K 8 way associative cache, don't try to warm up an 8K 16 way associative cache with it. 

I'm not sure what you mean by "inaccurate", I don't think there would be any inaccuracy. 

Polina

On Thu, Nov 19, 2009 at 1:33 PM, sparsh mittal ISU <sparsh@xxxxxxxxxxx> wrote:

Hello All
I wanted to ask regarding the following post of Mike Marty:

 load-caches will take a long time because it actually replays the contents

> of the caches as individual memory requests.  The reason we do this is so

> that we can use the same warmup files for different cache sizes and
> configurations (it approximates well).  After each load is completed, we
> clear the Ruby stats and continue the simulation.
> 


> I'm not positive how big our standard cache warmup files are, but I've
> seen it take anywhere between 10-45 minutes to load the caches.
> 

If we are only using cache warmup files for the same configuration, as we made checkpoint and cache file for, is there an alternative so that 10-45 minute time can be saved. Moreover, does anyone have any idea, how much inaccurate will the simulation be using these warm-up files? I would be thankful for the reply.
Thanks
Sparsh Mittal
ISU

_______________________________________________
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.