Parallel Slackness
•Valiant called the amount of “excess” parallelism
needed to cover latency parallel slackness
•In the
“best case” a parallel slackness of logP is required because in the best case
latency will be proportional to log P
•Any additional delays require
further slackness