Textbook Terminology (cont)
CC-NUMA: Cache Coherent - Non Uniform Memory Access architectures
- Clusters of shared-bus multiprocessors with an interconnection network between clusters
- Main memory physically distributed between clusters
- Cache coherence maintained within a cluster via a snooping protocol
- Cache coherence maintained between clusters via directory protocol.
- Main memory physically distributed among clusters - when accessing non-cache memory access time depends on distance to memory location.
- Differentiation made between processor memory, cluster shared memory, and global shared memory
- All processors can access global shared memory, cluster shared memory can only be accessed by processors within a cluster.
- How much memory of each type becomes a design issue
- Stanford DASH, Stanford FLASH, Wisconsin Multicube