Re: [Gems-users] A question about L2 wire delay

["lu peng" <penglu01@xxxxxxxxxxx>]

Alaa,

Thanks for your reply. For the NUCA setting, does RUBY implemented the L2 cache block migration according to visiting frequecy? If so, which part of source code links to this?

Thanks,

Lu

>From: Alaa Alameldeen <alaa@xxxxxxxxxxx>
>To: Gems Users <>
>CC: penglu01@xxxxxxxxxxx
>Subject: Re: [Gems-users] A question about L2 wire delay
>Date: Tue, 09 Aug 2005 09:03:29 -0500
>
>Lu,
>
>>I have a question about L2 wire delay in RUBY. If I have 128 L2
>>banks and 4 cores in one chip, what's the L2 banks distribution by
>>default? The delay of farther banks are bigger than those of nearby
>>banks. Does RUBY show this?
>
>You will have to create a network file that specifies how your cache
>banks are connected and the latencies of different links. Please see
>ruby/network/simple/Network_Files/ for examples.
>
>>In addition, I found there are 3 Latencies in rubyconfig.defaults:
>>L2_RESPONSE_LATENCY, L2_RECYCLE_LATENCY and L2_REQUEST_LATENCY.
>>Could somebody please tell me the differences?
>
>The usage of these parameters depends on the protocol. For the
>protocol that you are using, you can look for where these different
>parameters are used. But typically the names are self explanatory
>(e.g., L2_RESPONSE_LATENCY is added to a response from the L2 cache,
>...etc.).
>
>>One more suggestion: it would be better if there is a search
>>function in gems-users archives.
>
>That would be nice, but unfortunately our mailing list server
>doesn't support that feature yet! Thanks for the suggestion though.
>
>-Alaa