Hacker News Reader

I like consistent hashing (https://en.m.wikipedia.org/wiki/Consistent_hashing). When a hash table (or load balancing pool) needs to be resized, it usually reduces the number of keys (clients) that need to be remapped.

View on HN

Even simpler is Weighted Rendezvous Hashing (https://www.snia.org/sites/default/files/SDC15_presentations...).

It's quite a bit easier to implement and verify than consistent hashing, and carries the same benefits (minimal reshuffling on ring resizes, etc)

I recently updated the Python example on the Wikipedia page for Weighted Rendezvous Hashing. (I'm amazed my edit seems to be still there.)

https://en.wikipedia.org/wiki/Rendezvous_hashing

In a similar vein, Maglev hashing has a more uniform distribution and decent reshuffling. I couldn't get my head around the explainaions that I had seen until I read this one:

https://www.usenix.org/sites/default/files/conference/protec...

Reply on news.ycombinator.com