Re: [aqm] 答复: [iccrg] [tcpm] ECN support and usage on the Internet

[Mikael Abrahamsson <[e-mail]>]

On Tue, 19 Mar 2013, Jim Gettys wrote:

> I worry about bufferbloat in the HARQ layer of cellular networks not

This might be from just one vendor, but the Radio Link Control (RLC) layer 
on our equipment will buffer up to 400 packets before it does tail-drop. 
There might be additional buffering in GGSN/SGSN doing rate-shaping before 
it's even sent to the RNC (which terminates the RLC layer between the RNC 
and the end user terminal equipment). All this buffering is strict FIFO.

I personally belive that AQM at each of these buffering points would be 
beneficial to the end user equipment.

> My intuition is to get VOIP reliable over WiFI on a busy network (or 
> when far away from the AP) we'll want to classify the VOIP packets 
> anyway and use the right hardware class in WiFi; this is primarily to 
> get priority access to a transmit opportunity on WiFi rather than the 
> previous reasons we've had.  Certainly on other media we've seen little 
> reason to have to do any explicit classification for VOIP anymore once 
> running fq_codel.  Things generally "just work" without all that hair 
> for VOIP traffic.

I agree, fq_codel probably needs some DSCP handling so that VOIP packets 
(EF marked) would end up in a different hardware queue. For mobile, the 
bearer model would mean there would be an L4 inspection engine putting the 
traffic into the correct bearer anyway (that's the 3GPP way), so here 
codel probably wouldn't need to do anything (because there would be 
several codel instances running, one per bearer).

-- 
Mikael Abrahamsson    email: [e-mail]