Mult-level Caches with Split-Transaction Bus

• General structure uses queues between

  • Bus and L2 cache
  • L2 cache and L1 cache

• Deadlock!

• Classify all transactions

  • Request, only generates responses
  • Response, doesn’t generate any other transactions

• Requestor guarantees space for all responses

• Use Separate Request and Response queues

Previous slide

Next slide

Back to first slide

View graphic version