My thoughts here are:
(1) ICMP Source quench is more problematic in the general case than for
"wireless TCP" weenies, because most of the wireless TCP environments that
I've seen or heard about don't have lots of connections open. In the general
case, ICMP SQ can affect lots of connections that don't have any
relationship to the congestion being encountered. If you only have a few
connections open, this isn't as much of a problem.
Just because sending an ICMP SQ to www.yahoo.com is not a good thing doesn't
mean that sending an ICMP SQ to a wireless TCP host is a bad thing.
The other knock on ICMP SQ is, of course, that these messages aren't
authenticated, so that denial of service attacks are difficult to defend
against.
(2) Gab Montenegro and I had hoped that ECN would turn out to be an
unambiguous differentiation between congestion/loss and corruption/loss (if
you are seeing CE-bits set, slow down, if not, retransmit aggressively). The
ECN draft dispels this trip to networking fairyland by pointing out that
EC-bit-set packets are as likely to be dropped as any other during periods
of high congestion. So CE-bit set does ambiguously mean "slow down", but the
absence of CE-bit set means either "things are fine" or "things are
unimaginably bad".
So my take is that deploying ECN is a good thing for "long thin networks",
but not because ECN provides an unambiguous indicator of corruption loss.
This view is pointed out in the ECN section of
http://www.ietf.org/internet-drafts/draft-montenegro-pilc-ltn-00.txt.
This isn't intended as a criticism of ECN - what else could a router do,
when it runs out of buffers - and it isn't intended to imply that anyone
else had this interpretation, only to head off anyone who does!
Spencer
> -----Original Message-----
> From: Reiner Ludwig [SMTP:rludwig@huginn.CS.Berkeley.EDU]
> Sent: Thursday, January 21, 1999 8:04 PM
> To: Phil Karn
> Subject: Re: PILC: prioritization
>
> Phil,
>
> At 04:01 PM 1/21/99 , Phil Karn wrote:
> >
> >Right, a classic problem on dialup links. Both there and on CDMA I
> >experimented with ICMP Source Quench messages to keep the link queue
> >from getting too deep. I was mainly concerned about the latency
> >encountered by an interactive session sharing the link with a bulk
> >transfer, though of course TOS queuing is the better approach. In any
> >event, ICMP SQ messages worked very well at keeping the queues from
> >getting too big with no deleterious effect on throughput. I've never
> >accepted the conventional wisdom that SQ is bad.
>
> Shouldn't we much rather promote the use of the ECN scheme instead?
This archive was generated by hypermail 2b29 : Mon Jan 28 2002 - 09:12:19 EST