Cache's inherently point to NUMA Designs but I am just curious to know if there have ever been cache's designed with Uniform Memory Access patterns.
This comment was marked helpful 0 times.
dtaveras
@shabnam I think that NUMA refers to the design of the whole system not necessarily the cache design. Its hard design a system with uniform memory access because of the tradeoffs between the costs of the resources. Caches are typically implemented using SRAM but it would not make sense to use SRAM for the other memory resources since it is quite expensive as compared to DRAM or Disk. With the emergence of new memory technology such as PCM (Phase Change memory) we could see a memory system with a more uniform Memory Access Latency.
This comment was marked helpful 0 times.
tianyih
@shabnam Well I think the previous slide shows a great example of uniform memory access, since no core has its own local cache, all the accesses would have the same latency. Isn't it what we define as 'uniform'?
Cache's inherently point to NUMA Designs but I am just curious to know if there have ever been cache's designed with Uniform Memory Access patterns.
This comment was marked helpful 0 times.
@shabnam I think that NUMA refers to the design of the whole system not necessarily the cache design. Its hard design a system with uniform memory access because of the tradeoffs between the costs of the resources. Caches are typically implemented using SRAM but it would not make sense to use SRAM for the other memory resources since it is quite expensive as compared to DRAM or Disk. With the emergence of new memory technology such as PCM (Phase Change memory) we could see a memory system with a more uniform Memory Access Latency.
This comment was marked helpful 0 times.
@shabnam Well I think the previous slide shows a great example of uniform memory access, since no core has its own local cache, all the accesses would have the same latency. Isn't it what we define as 'uniform'?
This comment was marked helpful 0 times.