[Tech] Backoff considered harmful?

[Michael Rogers]

Ken Snider wrote:
> Am I too simplistic to think that the throttling system should mind a 
> semireliable way to guage delay due to Outgoing BW delay, divvy that 
> into number of outbound reqs/sec, and NAK any inbound requests that come 
> in above that rate?

I believe Toad's already implemented something like this, but based on 
estimated outgoing b/w usage rather than b/w delay. Hopefully with token 
passing it will neve be necessary to NAK requests, because we won't hand 
out a token until we're ready to accept a request.

Cheers,
Michael