Are there any tradeoffs in maintaining separate request/response queues? Would it require more space or is each response/request queue simply smaller?
Abandon
I am wondering where does the queue comes from? Is it a part of L1 or L2? And if the queue length can be changed. Where is the queue space comes from when we intend to increase the queue length? And how would the reduced queue space be utilized when we reduce queue length? I don't quite understand the hardware implementation of this queue design.
paracon
The size of the response queue could be as small as 1 since it will be immediately consumed. But it could lead to some stalling.
sandeep6189
@paracon why would there be stalling? Does responses have any dependency to fulfil?
zale
If the queue has length 1, the queue fills up immediately so a cache must take every response out of the queue immediately to avoid a risk of stalling the sender. This could become a problem when the receiver busy making requests and can't service responses right away.
Are there any tradeoffs in maintaining separate request/response queues? Would it require more space or is each response/request queue simply smaller?
I am wondering where does the queue comes from? Is it a part of L1 or L2? And if the queue length can be changed. Where is the queue space comes from when we intend to increase the queue length? And how would the reduced queue space be utilized when we reduce queue length? I don't quite understand the hardware implementation of this queue design.
The size of the response queue could be as small as 1 since it will be immediately consumed. But it could lead to some stalling.
@paracon why would there be stalling? Does responses have any dependency to fulfil?
If the queue has length 1, the queue fills up immediately so a cache must take every response out of the queue immediately to avoid a risk of stalling the sender. This could become a problem when the receiver busy making requests and can't service responses right away.