Good point. So the queue will be somewhat more full when the first drop
happens. The question is will it be full enough to ensure that the cwnd
after the drop is detected remains big enough to keep the pipe filled.
Some math or experimentation is needed to figure this out. I suspect
there will still be some situations where the AQM hurts single TCP
throughput, at larger RTTs. There are some tweaks I can think of to
avoid this.
Simon
On 5/22/2015 11:42 PM, Jonathan Morton wrote:
In practice, I haven't noticed any loss of throughput due to using
Codel on 100ms+ RTTs. Probably most servers now use CUBIC, which
contributes to that impression. Using ECN rather than tail drops also
makes the delivery smoother.
There is a flaw in your analysis. Codel only starts dropping (or
marking) when the sojourn time has remained above target (5ms) for an
entire interval (100ms), during which time the cwnd is still growing.
Thus the peak queue occupancy is more than 5ms.
It is however probably fair to say that a single Reno flow does lose
some throughput under AQM.
- Jonathan Morton
_______________________________________________
aqm mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/aqm
