[Tech] Re: Backoff considered harmful?

[Michael Rogers]

toad wrote:
> Hmm. Okay can you explain a little more about why we try a randomly
> chosen bucket first?

To achieve fairness when the buckets aren't full. Or rather, fairness on 
average - strictly speaking round-robin would be fairer, but it also 
reveals more information - each time you get a token from a peer you 
know that exactly n of that peer's searches have completed, whereas if 
buckets are chosen at random you know the average completion rate, but 
you don't learn anything about specific searches.

But that's probably a bit paranoid... round-robin should be fine.

Cheers,
Michael