Re: [mpls] [spring] redux: Special purpose labels in draft-hegde-spring-traffic-accounting-for-sr-paths

[<[e-mail]>]

John,

that’s not what I’m looking for. What I’m looking for is traffic statistics collected at transit nodes. These statistics should reveal the true end-to-end traffic demand within the MPLS domain. The collection of statistics shouldn’t add complexity. A low number of counters helps to simplify collection and post-processing.

If I’m not entirely wrong, this discussion at least partially discusses different ways how to define what kind of flow to account and where and how to capture related statistics. In that case I prefer a collection of a broader set of requirements and possible solutions. If our ongoing discussion doesn’t identify a set of useful options (I’m not sure whether all contributors to the discussion require the same here), a draft collecting different requirements and solutions may be helpful.

Regards,

Ruediger

Von: John E Drake [mailto:[e-mail]]
Gesendet: Donnerstag, 16. November 2017 13:00
An: Geib, Rüdiger <[e-mail]>; [e-mail]
Cc: [e-mail]; [e-mail]; [e-mail]; [e-mail]; [e-mail]
Betreff: RE: [mpls] [spring] redux: Special purpose labels in draft-hegde-spring-traffic-accounting-for-sr-paths

Ruediger,

There is also the possibility of using a GAL w/ a new fixed size GACH containing the SR Segment List Id.  This is similar to Robert’s suggestion of using a VXLAN header.

Yours Irrespectively,

John

From: mpls [mailto:[e-mail]] On Behalf Of [e-mail]<mailto:[e-mail]>
Sent: Thursday, November 16, 2017 4:44 AM
To: [e-mail]<mailto:[e-mail]>
Cc: [e-mail]<mailto:[e-mail]>; [e-mail]<mailto:[e-mail]>; [e-mail]<mailto:[e-mail]>; [e-mail]<mailto:[e-mail]>; [e-mail]<mailto:[e-mail]>
Subject: Re: [mpls] [spring] redux: Special purpose labels in draft-hegde-spring-traffic-accounting-for-sr-paths

Adrian,

to me, there’s no ideal solution. But an analysis may help to find a useful solution. There’s a need to collect traffic statistics also for packets which don’t follow the shortest end to end path. There’s no simple howto, I think.

For the time being, I’d prefer not to add special labels to the stack. What other options are there?

  *   Accounting at the router pushing a relevant label stack only.
  *   Accounting of an n-label stack.
  *   Acoounting of a subset of labels only (e.g. Node-SID Labels and Anycast-SID, but not ADJ-SID). The idea is a compromise to limit the number of counters be maintained. Consider accounting of the top 2 labels carrying global routing information.
  *   A special label. Shradda proposes to put such a label into the stack. The labels present there prior to the addition are maintained. One might think about a single top label which identifies and replaces the label stack carrying routing information relevant for the path. That would simplify accounting, but it requires suitable IGP functionality.

None of the options sounds simple. Are there more (and simpler) ones I didn’t come upon?

Regards, Ruediger

Von: spring [mailto:[e-mail]] Im Auftrag von Adrian Farrel
Gesendet: Donnerstag, 16. November 2017 06:35
An: 'Mach Chen' <[e-mail]<mailto:[e-mail]>>; 'Jeff Tantsura' <[e-mail]<mailto:[e-mail]>>; 'Robert Raszuk' <[e-mail]<mailto:[e-mail]>>
Cc: 'draft-hegde-spring-traffic-accounting-for-sr-paths' <[e-mail]<mailto:[e-mail]>>; 'spring' <[e-mail]<mailto:[e-mail]>>; 'Zafar Ali (zali)' <[e-mail]<mailto:[e-mail]>>; 'mpls' <[e-mail]<mailto:[e-mail]>>
Betreff: Re: [spring] [mpls] redux: Special purpose labels in draft-hegde-spring-traffic-accounting-for-sr-paths

Let's unpick a couple of things...

1. This work is not talking about per-flow accounting, it is talking about peer SR-path accounting
2. ipfix on its own does not cut it because you still have to put a marker in the packets
3. Yes, SR assumes there is no (i.e. zero) state per SR-path in the network
But this third point causes a tension: we want to use SR because it is good, but we want to do transit node diagnostics because (frankly) they are necessary.
To get the full picture of why they are necessary read the draft, or consider ECMP.

This discussion will not be unfamiliar to those who tried to debug LDP networks.

Adrian