Re: [re-ECN] RE-ECN without ECN.

["Don Bowman" <[e-mail]>]

On Mon, Oct 5, 2009 at 5.45 PM, Matt Mathis wrote:
> On Mon, Oct 5, 2009 at 3:03 PM, Fred Baker <[e-mail]> wrote:
> >
> > On Oct 1, 2009, at 9:36 AM, Matt Mathis wrote:
> >
> >> If you add the assumption that the network itself does not reorder
> >> packets, then the number of out-of-order packets is a direct measure
> >> of the upstream congestion.  (e.g. the first transmission was lost,
> >> and I see the retransmission).
> >
> > The assumption concerns me. There are probably places in the Internet
> where
> > loss is unusual and reordering is unusual, but making it a baseline
> > assumption... The Internet is still a best effort service, which is
> to say
> > that datagrams may be lost, duplicated, or reordered.

One thing to watch out for is hand-off conditions. E.g. in a mobile
network, handing from one cell to another, or one carrier to another,
can cause some re-ordering w/o loss. Also, periodically you run into
a network which balances by packet rather than flow, or balances
by flow on something which isn't stable and flops back and forth.
Typically OSPF ECMP does a L3(or L4) hash to select a link, but sometimes
that hash is of an 'outer' ip which is not relevant to the inner content,
and thus reordering occurs.

... snip ...

> 
> I agree all around.   Bob has been putting a lot of effort into making
> sure that the specifications for tunnels say the right things.   What
> I don't know is if there have been any measurements of the population
> of devices that do busted things, such not updating the inner ECN bits
> from the outer header on decapsulation.   (But note that this is
> busted for all variants of ECN).

Again to the mobile network. One of the most (variably) congested networks
out there is the mobile backhaul network. In 3GPP, this is carried in a
tunnel type called GTP-U. I've not found GTP-U implementations to be
much for moving markings from outer to inner (e.g. DSCP). I've done
a quick check on a couple of networks, and did not observe ECN marked
on the inner, so i can't say if they support ECN 'moving the mark'.

One of the things i am hopeful to use re-ECN for in our products is
as an indication of backhaul health (and i dream someday of RF health)
so that we can break reporting information down by 'this is what
congested/near-congested/non-congested links are doing'. So i'm
hopeful it can be made to work in GTP-U (which implies getting
ECN basis working).

--don