Re: [tsvwg] L4S DSCP (was: L4S drafts: Next Steps)

[Jonathan Morton <[e-mail]>]

> On 28 Mar, 2021, at 11:03 pm, Ingemar Johansson S <[e-mail]> wrote:
> 
> And these that "can or will not.." will then continue to run across the internet 10 years from now ?, 20 years ?

Yes.  Exactly.

I mentioned before that Pete's Czech ISP is still running 15-year-old routers that date from their founding.  I personally still have a "UFO" Apple Airport Base Station - the original one with 802.11b, 56-bit WEP, and a 56K analogue modem - which doesn't get much *use* these days, but it still *works* and is potentially useful.  Heck, I still write code for 40-year-old computers as a hobby.

These old devices lack support for newer specifications.  But they still work just fine in an environment that does use them, because backwards compatibility was designed in when the new specifications were published.  ECN Capable transports still work with dumb drop-tail FIFOs, and Ethernet frames carrying IPv6 packets still traverse Layer 2 devices that were in the shops before IPv6 was finalised, and which require IPv4 (or maybe AppleTalk) packets to configure the settings.  That is how Internet engineering works.  In general, old network devices are expected to continue working as before, and you only need to upgrade if you want to benefit from the new technology.

But L4S is *not* designed that way.  If it is deployed as-is without effective guarding, it will break quite a lot of stuff that isn't even very old.  Your best proposal is to scrap or materially change an existing specification that is currently rapidly deploying in the Internet because people see real benefits from it.  It's entirely conceivable that Pete's Czech ISP will choose this year to finally upgrade those old routers, *specifically* for the purpose of deploying fq_codel, now that they have some experience with it - and that would start a new 15-year timer, wouldn't it?

I've personally put forward two different ways to make L4S safely deployable in the past week - one which requires only minimal technical changes to at least try out, and one which changes much more but which I think will work better overall.  As before, it looks like those suggestions are being rejected out of hand (though Koen's attitude appears to be more balanced, for a change).  This is precisely what has delayed L4S for two years and counting, and there is no end in sight.

 - Jonathan Morton