Re: [tsvwg] CC/bleaching thoughts for draft-ietf-tsvwg-le-phb-04

[<[e-mail]>]

Hi Mikael,

my views marked [RG].

Regards, Ruediger

-----Ursprüngliche Nachricht-----
Von: tsvwg [mailto:[e-mail]] Im Auftrag von Mikael Abrahamsson
Gesendet: Donnerstag, 19. April 2018 08:38
An: [e-mail]
Betreff: Re: [tsvwg] CC/bleaching thoughts for draft-ietf-tsvwg-le-phb-04

On Thu, 12 Apr 2018, [e-mail] wrote:

> advantage ("prove", not "believe").. Someone has to configure and 
> operate all this scheduling, and sometimes debugging is required. If 
> something is new, a basic standard set which is expandable to me is 
> more favorable, than a flexible (aka complex) toolset.

I've had some discussions with other people in the operational community, and multiple questions have come up.

1. If I do not bleach at all, should I do something based on this new draft? (there are people with MPLS networks that do their "bleaching" by assigning all traffic to BE on their transport (802.1p=0 / MPLS EXP=0), so they actually do not bleach customer diffserv in the IP header.

[RG] There's no obligation to bleach. The current standards allow bleaching if a domain wants to protect internal 
DiffServ transport from traffic which is received with Diffserv markings without having been negotiated. One reason is, 
that in that case, the meaning of the DSCPs is simply unknown to the receiver.

[RG] Its perfect to be liberal with what you receive - the conservative part is sending. Let's assume my domain is following on yours. My OP-folk receives packets marked with a DSCP from your domain and asks, which PHBs are linked to these DSCPs? An honest response would be "dunno".

2. If I do bleach, should I remap CS1 to the new LE CP?

[RG] Only if the sender advised you, that packets marked by CS1 indeed classify for LE! We've discussed this more than once: DSCPs do not carry any end to end information. A DSCP which is received from a neighbor domain, which is not bothering to inform you about the PHB linked to their DSCP markings, to your domain is an unknown DSCP. Your staff may degrade traffic to LE by the remap action you describe.

[RG] I'm not sure whether we've checked that all domains sending CS1 today actually classify it as LE.

[RG] The carrier I work for supports Diffserv.

.........

So let's start with the operator that does no bleaching:

1. PHB should be that CS1 and LE is treated the same, ie lower priority than BE.
2. Do not bleach from customer.
3. Re-map CS1 to LE at edge towards other ISPs by default, but ISPs can agree on different behaviour.

[RG] As above, don't agree as long as I don't know whether you assume the PHB linked to a DSCP to be known and agreed by both interconnected parties.

For the ones that do bleaching:

>From customer/other ISP.
1. Remap CS1 to LE.
2. Do not bleach LE.
3. Bleach rest to BE.
4. PHB behaviour should be LE gets lower priority than BE.

[RG] As above, don't agree as long as I don't know whether you assume the PHB linked to a DSCP to be known and agreed by both interconnected parties. Further, a remap rather than bleaching means that there's an agreement on the PHB linked to CS1 between the interconnected networks. Correct?

[RG] You only discuss the case where two interconnected networks both support LE, correct?

Thoughts?

-- 
Mikael Abrahamsson    email: [e-mail]