(3.2.5) Token Coherence

Milo M. K. Martin, Mark D. Hill, and David A. Wood, "Token Coherence: Decoupling Performance and Correctness," International Symposium on Computer Architecture (ISCA), June 2003. IEEE Xplore Link

Technology trend
     workload (good thread level parallelism, low latency preferred) > snooping
     Technology trend > directory

+ Fast cache to cache misses
+ No bus like interconnect
     avoid virtual bus ordering
+ Bandwidth efficiency

Token counting > ensures safety
Persistent requests > prevent starvation
Decoupling correctness and performance

Broadcast with direct responses
use unordered interconnect

Need one token to read, all tokens to write.
To avoid data sent with all tokens : owner token (clean/dirty)

- Non-silent eviction 

Prevent starvation
     invoke after a timeout
     send to all componenets
     all comp remembers in a table and continually redirects all tokens to requestor
     Deactivate when complete
What if many processors issue persistent req simulat?
     Use starvation-free arbiter
     Single/Banked/Distributed
need to prevent reordering of activate and deactive messages

- Scalability of persistent requests

Predictor based (reduce broadcasts)
     broadcast-if-shared. 
     Owner
     Groups