Re: [iccrg] [tsvwg] IETF104 ICCRG TCP Prague presentation

[Jonathan Morton <[e-mail]>]

> On 4 Apr, 2019, at 12:40 pm, Greg White <[e-mail]> wrote:
> 
> I don't doubt that the SCE signaling method could be used in a variant of DCTCP, but the *huge* downside is that it can only work on bottleneck links that are able (or willing) to support FQ, which will be precious few.

Let me stop you right there.  You are remembering a discussion we had at least two weeks ago, based on an early concept of SCE which has since been superseded.

We have running code which converges rapidly to fairly sharing the link between two SCE flows, with only a single-queue SCE-aware AQM.  That is, we used the "besteffort" and "flowblind" modes of Cake, which disable its Diffserv and flow-isolation capabilities, specifically to test that aspect of our work.  Since our prototype SCE response is DCTCP-like, that result should not be a surprise to you.

We *also* believe we have a solution for fairly sharing a single queue between SCE and non-SCE flows.  We haven't implemented and tested it yet, however.  Work continues.

> (unless you think blocking implementers from supporting dual-queue is some kind of perverse benefit)

Dual-queue AQM can still be supported through Diffserv, as I have repeatedly pointed out.  The associated PHB can specify the "short queue" and "no reordering" behaviours you're so keen on.

Now, reading back up your post:

> Has any progress been made on a variant of Cake that utilizes the L4S ECT(1) semantic?  It would be useful for comparison purposes.

We have not done so.  The code is freely available if you wish to try it yourself; it should not be difficult.  Cake also offers an excellent platform for exploring the behaviour of single-queue ECN-aware AQMs, which I think you would find valuable.

> I don't believe that the RFCs are on your side here.  RFC 3168 contemplates, and rejects, that definition of ECT(1)…

It also contemplates, rejects, and suggests a better alternative for L4S' use of ECT(1), right there in the quote.  And on the subject of RFC-8311, that document contains a section entitled "Effective Congestion Control Is Required", and it specifically leaves alone the several suggested uses of ECT(1) in its edits to RFC-3168.

I note in passing that SCE has the support of the TSVWG Chairs.  That is sufficient motivation to continue work to establish its true merits and limitations.

 - Jonathan Morton