Re: [Gems-users] Timing Model

[Mike Marty <mikem@xxxxxxxxxxx>]

>
> I would like to learn more about GEMS memory timing model when the
> MSI_MOSI_CMP_directory protocol is used.
>
>
>
> My first question maybe is quite banal:
>
> What is the latency unit? Is it the ruby cycle?
>

Yes, it is the Ruby_cycle.

>
>
> In the ruby configuration file I see the following timing parameters:
>
>
>
> -          L1_response_latency
>
> -          L2_response_latency
>
> -          L1_request_latency
>
> -          L2_request_latency
>
>
>
> What do response_latency and request_latency parameters refer?
>

These are parameters that may be used by a SLICC specification for a
protocol.  The SMP protocols typically use the CACHE_RESPONSE_LATENCY
rather than the separate L1 and L2 response latencies.

>
>
> Among the parameters I also see on_chip_link_latency and
> network_link_latency. I think that the first one is ignored when the
> file_specified network topology is used. Is it right?
>

They are both ignored when the file specified network is used.  In this
case, you specify the latency and bandwidth of each individual link.  The
bandwidth is multiplied by the bw_multiplier to specify thousanths of a
byte per cycle.

 >
>
> What does network link latency refer?
>
>
>
> Is cache timing model independent from the number of sets? If the right
> response is yes, it means that a parallel search is assumed. Is it right?
>

We do not dynamically calculate the cache latency based on configuration
parameters.  The cache latency uses one of the fixed parameters, that you
mentioned above.  It depends on what the SLICC controller uses.

Sorry for the late reply.  Many of us have been working on papers.

--Mike