On Thursday 01 June 2006 16:00, Ian Clarke wrote: > > On 1 Jun 2006, at 10:52, Matthew Toseland wrote: > > > Maybe we should have the initial backoff time dynamically calculated? > > Discuss. > > > > [13:30] <edt> toad_ simpiler idea use max(n*myping,targetNodes*ping)*2 > > as target nodes min window > > [13:30] <edt> toad_ simpiler idea use max(myping,targetNodes*ping) > > *2 as > > target nodes min window > > [13:31] * edt sighs > > [13:31] <edt> here is is correctly expessed > > [13:31] <edt> toad_ simpiler idea use > > max(myPingTime,targetNodesPingTime)*2 as target nodes min window > > Can you explain the rationale behind this formula?
My thoughs are like this. Now we have picked a number out of the air to use as a minimum window value (5 seconds). The ping time gives us the absolute fastest time we are going to get a reply from a given node. The *2 is a guess that, if a packet contains data, it will take longer to process - '2' may not be optimal... Why should we not use a minimum window that is based on how fast a given connection is working. Note this does not change the backoff logic, just allows it to send more packets (smaller window) IF the connection is fast enough. Ed
