> Is there a point where we just start using HPPC or do we do something it
doesn't and so the performance tests here are simply a pedagogical exercise
for those interested in hashing strategies

Hard to tell for sure without a realistic use case I think. Microbenchmarks
may be misleading, the general costs of whatever is computed may swallow
minor differences between implementations. Plus there *are* minor
differences between implementations:

1) resize/rehash strategy is different (prime vs. power-of-two),
2) slot lookup strategy is different (rehash-of-hash + linear lookup vs.
hash + double hashing).

If nobody invests the time to actually test over a large(r) body of
applications then I'd say the switch doesn't make much sense at this point.

Dawid

Reply via email to