Re: [tsvwg] Neal Cardwell's rationale for supporting ECT(1) as an input/L4S signal

[Steven Blake <[e-mail]>]

On Tue, 2020-05-12 at 23:38 +0300, Jonathan Morton wrote:
> > On 12 May, 2020, at 11:29 pm, Neal Cardwell <[e-mail]>
> > wrote:
> > 
> > > The shallow queues referred to are intended for deployment in
> > > tightly
> > > controlled datacentre environments, where the typical path RTT is
> > > measured in
> > > microseconds rather than milliseconds. It would not be surprising
> > > if a
> > > transport passing through such a queue as part of a typical
> > > Internet path of
> > > many milliseconds experienced poor utilisation.
> > 
> > I think the issue is that such paths do not currently suffer from
> > throughput problems, but could if they attempt to participate in an
> > SCE conversation.
> 
> Except that they *do* already suffer from throughput problems if they
> attempt to participate in an RFC-3168 ECN conversation, to precisely
> the same extent, and for precisely the same reasons.

In (the vast majority of) datacenter switches, all queues are shallow
(10s of microseconds) relative to typical Internet RTTs (tens of
milliseconds). It is completely orthogonal to any AQM or ECN scheme.

If you want your Internet-facing servers buried within a datacenter
network, you are well advised to carve out a BW capacity slice for that
traffic sufficient to avoid congestion.


Regards,

// Steve