XMind

Multiprocessors and Cache Coherence

1. Catagory

   1. SISD - uniprocessor

   2. SIMD - data-level parallelism

   3. MISD - no commercial product

   4. MIMD - thread-level parallelism

      1. Shared-memory

         1. SMP(UMA style)

         2. NUMA

      2. Cluster

         1. commodity clusters

         2. custom clusters(high end)

2. Cache Coherence

   1. coherence, defines what values can be returned by a read

   2. consistency, determines when a written value will be returned by a read

   3. protocols

      1. Snooping(good for UMA)

         1. bus

         2. scalability problem

      2. Directory(good for NUMA)

   4. Performance: coherence misses

   5. Implementing Locks

      1. atomic primitive -> lock in memory

      2. cache coherence + atomic primitive(double check) -> lock in cache

3. Memory Consistency

   1. sequential consistency

      1. speculation

   2. relaxed consistency

      1. relax W→R: total store ordering (processor consistency)

      2. relax W→W: partial store order

      3. relax R→W and R→R: weak ordering, the PowerPC consistency model, and release consistency