Overview
The CMP-NUCA protocol is based on the protocol described in [http://www.cs.wisc.edu/multifacet/papers/micro04_cmp_wire_delay.pdf]. CMP-NUCA supports static and dynamic mapping of cache blocks and allows cache blocks to migrate and replicate within the dyanmic mapped caches. CMP-NUCA utilizes token coherence and a distributed presistent request mechanism to maintain coherence.
Organizations
SNUCA
- Static placement of cache blocks
To Run
- g_NETWORK_TOPOLOGY: FILE_SPECIFIED g_DYNAMIC_TIMEOUT_ENABLED: false REMOVE_SINGLE_CYCLE_DCACHE_FAST_PATH: true g_CACHE_DESIGN: NUCACOL g_adaptive_routing: false NUMBER_OF_VIRTUAL_NETWORKS: 7 g_endpoint_bandwidth: 1000 g_NUCA_PREDICTOR_CONFIG: SNUCA
DNUCA
- Dynamic placement of cache blocks allowing blocks to migrate towards requesting processors
To Run
- (all the same parameters as SNUCA except the following) g_NUCA_PREDICTOR_CONFIG: DNUCA ENABLE_MIGRATION: true g_NUM_DNUCA_BANK_SETS: 2 (or 4 or 16 depending on the number of banks) COLLECTOR_HANDLES_OFF_CHIP_REQUESTS: true PERFECT_DNUCA_SEARCH: true // If perfect searches is desired
RNUCA
- Dynamic placement of cache blocks allowing blocks to migrate and replicate towards requesting processors
To Run
- (all the same parameters as SNUCA except the following) g_NUCA_PREDICTOR_CONFIG: RNUCA ENABLE_MIGRATION: true ENABLE_REPLICATION: true g_NUM_DNUCA_BANK_SETS: 2 (or 4 or 16 depending on the number of banks) COLLECTOR_HANDLES_OFF_CHIP_REQUESTS: true PERFECT_DNUCA_SEARCH: true // If perfect searches is desired
Limitations
- Only an eight processor single CMP configuration is supported. The L2 cache must be split into 32, 64, or 256 banks. No other L2 bank totals are supported.