Re: [tsvwg] severity of interaction between L4S and non-L4S flows (was: draft-ietf-tsvwg-l4sops-01.txt)

[Pete Heist <[e-mail]>]

Regarding the sub 10 seconds guidance topic that came up in Monday's
meeting, there wasn't much response when the question was raised what
the guidance should be, probably because it's complicated, and depends
upon, at a minimum, RTT, flow lengths, the exact CCAs and AQM in use,
what metric matters to the application, etc. As we have shown in our
FCT data linked to earlier in this thread, the harm in 3168 queues
isn't limited to rate unfairness.

In the interest of keeping things simple, what about just removing that
bullet?

Otherwise, we might want to perform more extensive FCT tests with a
range of flow lengths, RTTs, bandwidths and CCAs, and I'm not sure if
anyone has the time to test this at the moment.

Pete

On Tue, 2021-07-13 at 12:16 +0200, Pete Heist wrote:
> Hi Greg,
> 
> Regarding the guidance for general purpose servers in Section 4:
> 
> ---
> *  Taking action based on the detection of RFC3168 FIFO
>    bottlenecks is likely not needed for short transactional
>    transfers (e.g. sub 10 seconds) since these are unlikely to
>    achieve the steady-state conditions where unfairness has been
>    observed.
> ---
> 
> Do you have any data to back up the sub 10 seconds threshold?
> 
> What we've seen is non-L4S flows yielding quickly to L4S flows in
> RFC3168 queues. In this plot of Prague vs CUBIC in a single Codel
> queue, the flows start at T=5 seconds, harm to CUBIC occurs almost
> immediately, and CUBIC's steady state near minimum cwnd is reached 4-
> 5
> seconds after flow start:
> 
> https://sce.dnsmgr.net/results/l4s-2020-11-11T120000-final/l4s-s6-rfc3168-1q/l4s-s6-rfc3168-1q-ns-prague-vs-cubic-noecn-fq_codel-50Mbit-20ms_tcp_delivery_with_rtt.svg
> 
> We also posted a test showing the FCT harm to short flows by a single
> long flow, where flow lengths for the short flows were taken from a
> lognormal distribution with a P5 of 64KB and a P95 of 2MB:
> 
> https://github.com/heistp/ccafct#sample-output
> 
> Thanks,
> Pete
> 
> On Mon, 2021-07-12 at 16:01 -0700, [e-mail] wrote:
> > 
> > A New Internet-Draft is available from the on-line Internet-Drafts
> > directories.
> > This draft is a work item of the Transport Area Working Group WG of
> > the
> > IETF.
> > 
> >         Title           : Operational Guidance for Deployment of
> > L4S in
> > the Internet
> >         Author          : Greg White
> >         Filename        : draft-ietf-tsvwg-l4sops-01.txt
> >         Pages           : 21
> >         Date            : 2021-07-12
> > 
> > Abstract:
> >    This document is intended to provide guidance in order to ensure
> >    successful deployment of Low Latency Low Loss Scalable
> > throughput
> >    (L4S) in the Internet.  Other L4S documents provide guidance for
> >    running an L4S experiment, but this document is focused solely
> > on
> >    potential interactions between L4S flows and flows using the
> > original
> >    ('Classic') ECN over a Classic ECN bottleneck link.  The
> > document
> >    discusses the potential outcomes of these interactions,
> > describes
> >    mechanisms to detect the presence of Classic ECN bottlenecks,
> > and
> >    identifies opportunities to prevent and/or detect and resolve
> >    fairness problems in such networks.  This guidance is aimed at
> >    operators of end-systems, operators of networks, and
> > researchers.
> > 
> > 
> > The IETF datatracker status page for this draft is:
> > https://datatracker.ietf.org/doc/draft-ietf-tsvwg-l4sops/
> > 
> > There is also an HTML version available at:
> > https://www.ietf.org/archive/id/draft-ietf-tsvwg-l4sops-01.html
> > 
> > A diff from the previous version is available at:
> > https://www.ietf.org/rfcdiff?url2=draft-ietf-tsvwg-l4sops-01
> > 
> > 
> > Internet-Drafts are also available by anonymous FTP at:
> > ftp://ftp.ietf.org/internet-drafts/
> > 
> > 
> 
>