Evaluation studies Min & Baers study
Results
miss ratio
- similar for both compared schemes, regardless of the number of processors
- considerably higher for randomly distributed iterations
write traffic
- higher for the SW scheme (write-backs of dirty lines at the end of each instance of one epoch)
- independent of the number of processors and process scheduling policy
network traffic
- higher for the HW scheme (multicasting for inductive invalidation)
- for the HW scheme, rapidly increases with the increase of the number of processors (SW scheme -
self-invalidation)
- write traffic and misses have a relatively small impact on the network traffic