Re: [tsvwg] Fwd: New Version Notification for draft-heist-tsvwg-ecn-deployment-observations-00.txt

[Pete Heist <[e-mail]>]

On Mon, 2021-03-01 at 07:54 -0800, Dave Taht wrote:
> I had pinpointed a whole bunch of mismarked ecn traffic over
> bittorrent to an ISP in argentina last year. In fact, all their
> traffic was mismarked. I can find the relevant IPs...

Just checking- meaning all their BT traffic was mismarked?

I only looked at one non-libtorrent client, Transmission, which seemed
to not set any bits in the DS field.

> On Fri, Feb 26, 2021 at 3:58 PM Jonathan Morton <
> [e-mail]> wrote:
> > 
> > > On 27 Feb, 2021, at 1:31 am, Martin Duke
> > > <[e-mail]> wrote:
> > > 
> > > I wonder if the non-TCP ECN traffic is QUIC? I don't think the
> > > main implementations are doing it, but there are a few ECN-
> > > capable implementations in production.
> > 
> > This would be a reasonable interpretation of traffic on ports
> > typically used with HTTP or similar L5 protocols.  We have,
> > similarly, interpreted non-TCP traffic using ports commonly
> > associated with tunnel protocols as actually being (at least
> > partly) TCP through a tunnel.
> > 
> > However, most of what we saw was on random high ports, and some
> > that are most often used by BitTorrent clients.  So we think it's
> > probably due to BT clients trying to do "clever" things with
> > TOS/DSCP but misunderstanding it.  I believe the default in
> > libtorrent is to set CS1, correctly, so I don't know which
> > particular clients might be so affected, or whether it is due to
> > end users setting manual firewall mangle rules incorrectly.
> > 
> > Since the data collection tool we used doesn't have a way to
> > unambiguously discriminate between QUIC and any other UDP-supported
> > protocol, it's difficult for us to say more about how much QUIC-ECN
> > was seen.  If you have a criterion we could use for that, we might
> > be able to build it into a future run.
> > 
> >  - Jonathan Morton
> 
> 
>