> I dont understand the meaning of these parameters and what queue/structures they represent in the system
>
I strongly suggest grepping through the code or using your favorite C/C++
cross-referencing utility. Many of the parameters are only used for
certain configurations and protocols.
> L2_RESPONSE_LATENCY
Used by some protocols (MOSI_MSI_CMP_directory,MOESI_CMP_directory, etc)
> RECYCLE_LATENCY:
Some protocols "recycle" a MessageBuffer. This is popping the message off
of a queue and re-enqueing it with the above latency. It essentially
simulates non-FIFO queues. Parameter is used in
$GEMS/ruby/buffers/MessageBuffer.C
> L2_RECYCLE_LATENCY:
> COPY_HEAD_LATENCY
Sorry, I don't recall what these are used for off the top of my head.
> SEQUENCER_TO_CONTROLLER_LATENCY
>
The latency incurred to reach a cache controller. In some SMP
configurations, this actually corresponds to the L2 hit latency (the SMP
protocols where each controller models both the L1 and L2 caches)
> If i am providing the network file why do I need to specify
> NETWORK_LINK_LATENCY
> ON_CHIP_LINK_LATENCY
>
You don't if you are providing a network file.
> How ruby multiplier affects simulation timing ? If we are using ruby with simics
>
> g_think_time: 5
> g_hold_time: 5
> g_wait_time: 5
>
These are only used by one of the Ruby tester micro-brenchmarks. These
parameters can be ignored.
--Mike
|