Matthew Toseland wrote: > But if the tokens are separate > from the queue, how is the queue size limited? As far as I can see it is > necessary that the maximum token bucket size *include* queued/running > requests as well as possible ones.
A request is only added to a queue after a token's removed from a bucket, and a token's only added to a bucket after a request is removed from a queue, so (total bucket size + total queue size) is constant. Do we need a further restriction on the number of queue slots each peer can occupy, given that the number of tokens it receives is already controlled? Cheers, Michael
