Go backward to Memory Access Go up to Top Go forward to NUMA Architectures

Distributed Shared Memory

• Logically, a global address space is provided.
• Physically, the memory is distributed among the processors.
• Remote memory accesses via a scalable network.
• NUMA architectures
  • Non-Uniform Memory Access
  • Local memory can be accessed faster.
• COMA architectures
  • Cache Only Memory Achitecture.
  • All memory transformed into caches.