-1

For algorithms such as the gossip protocol used by Cassandra, what is the advantage of random peer selection versus a deterministic approach such as iterating over the list of all peers?

Coder
  • 441
  • 2
  • 17
  • Related question: http://stackoverflow.com/questions/27077701/how-cassandra-select-the-node-to-send-request – Aaron Feb 22 '15 at 16:13
  • @BryceAtNetwork23 That question is only related in that gossip is mentioned as a source of data. This question pertains to why the anti-entropy gossip implementation in Cassandra has uniform random peer selection versus a deterministic approach. – Coder Feb 23 '15 at 00:39

1 Answers1

0

At a guess, I'd say it equalizes (statistically speaking) the time taken to detect failure for all peers in the list. If it were to iterate over a list of peers, then a failure in a peer that is near the tail of that list takes much longer to detect than one in a peer that's at the head of the list. I want to say something about better average case performance across the board, but I'm going to leave that to someone more experienced.

AmorphousCrystal
  • 64
  • 6