While reviewing PIE, it might be worth revisiting my comment here: https://mailarchive.ietf.org/arch/msg/aqm/ifkBe6uvME5Nqm-J6ulZWlooBxo The second part was never answered to my satisfaction.
I believed I had found a bug: I said, "I think there are a variety of traffic patterns that can allow 'p' to meander around in non-zero space even though the queue is nearly empty" Maybe you've addressed this, because I think it was due to a heuristic. I'm much in favor of removing the heuristics on control systems; thanks. -Dave -----Original Message----- From: aqm [mailto:[email protected]] On Behalf Of Bob Briscoe Sent: Thursday, March 30, 2017 2:47 AM To: Jonathan Morton; Rong Pan (ropan) Cc: Greg White; tsvwg IETF list; AQM IETF list; [email protected] Subject: Re: [aqm] Questioning each PIE heuristic Jonathan, Picking up on an earlier point you made about avoiding heuristics by ensuring the underlying algo is sound,... that's precisely why I'm going through all the (9) PIE heuristics... For PI2 we removed all but 2 and it worked the same or better than PIE in all our tests. I have been assessing each of the other 7 one by one for reinstatement. So far I've rejected 6. I think I can reject this last one by making the sampling time of the base PI algo dependent on the max link rate. Then when the queue goes idle, the base PI algo will decay drop down to zero no slower than the queue drains, without needing this extra heuristic. But I need to check that's realistic. We will be writing all this up (probably in an update to the PI2 paper - I don't think the IETF PI2 spec is the right place for a critique of heuristics that it doesn't use). Our aim is a completely sound AQM in a few lines of code and a few operations so it can be implemented everywhere with minimal resistance from developers due to performance concerns (e.g. cheap ethernet switches, cheap home gateways, carrier-grade equipment for thousands of users, etc). Bob On 28/03/17 07:25, Jonathan Morton wrote: > > By all means, avoid dropping packets when the queue is actually empty - that > is, when you’re delivering the last packet in the queue. In that case, there > is no congestion to signal for. But there really is no need to have any > complex state-switching logic for that. If your underlying algorithm is > sound, it will naturally decay to zero packet drops if the empty-queue > condition persists. > > - Jonathan Morton > -- ________________________________________________________________ Bob Briscoe http://bobbriscoe.net/ _______________________________________________ aqm mailing list [email protected] https://www.ietf.org/mailman/listinfo/aqm _______________________________________________ aqm mailing list [email protected] https://www.ietf.org/mailman/listinfo/aqm
