[spring] measurement requirements (was Re: [mpls] Special purpose labels in draft-hegde-spring-traffic-accounting-for-sr-paths)

[Martin Horneffer <[e-mail]>]

Hi,

first thank you Shraddha for bringing the topic of traffic measurement 
to the lists.
And thanks to Stephane for focusing on the - from my point of view - 
most important aspects.

Apparently you can have different requirement for traffic measurement 
and based on those you'll need more or less complex solutions. And then 
you can often place different degree of complexity and different points.

While I don't know everybody's measurement requirements I can name those 
from an operator of a large multi protocol backbone that bears many 
different services on a global scale. Tools for network simulation and 
optimization like those mentioned by Robert - but not limited to that 
particular tool - appear to have the same basic requirement:

  - On a given topology, which can be the whole MPLS cloud, or as well a 
"core" like subtopology, you need an end-to-end traffic matrix in order 
to be able to simulate the exact traffic distribution to that topology 
and for possible changes, be it network remodeling or traffic 
engineering optimizations.

  - At least for our network the design goal was always to keep 
complexity for the forwarding nodes as low as possible. Neccessary 
complexity should better be moved to external tools that do the traffic 
matric calculation and network simulation. Even though we always had a 
strong focus on traffic measurement we never wanted to change the actual 
forwarding just for the sake of measurement.

What this really means is that we don't need to measure every single 
path an every node. Thus the current proposal is quite over-engineered 
for our purpose (and probably overly complex).

If we look at current MPLS-LDP networks, contemporary devices have 
enough means for ende-to-end traffic matrix measurements. Regardless of 
whether they are called "MPLS counters" or "LDP traffic statistics": 
there are per-FEC counters and they are well-suited, when taken from 
alle involved routers, to exactly get the traffic matrix. And there are 
usually counters at head ends of traffic enigneering tunnels of any 
kind, so that you always can verify the effect of related traffic 
engineering measures.

Per-FEC counters do not require additinal state in routers. They merely 
require a counter for every FIB entry they need to have anyways.

Do we need a requirements document to get the same for SR?

Other ideas for getting the traffic matrix are welcome, just please 
don't add more complexity on the forwarding machinery, e.g. don't add 
more labels than really needed.

Best regards, Martin


Am 16.11.17 um 11:48 schrieb [e-mail]:
>
> Hi,
>
> Yes today we do not have any CLI command on any router to get paths 
> statistics for LDP (I mean Ingress to Egress) as LDP is based on MP2P 
> LSPs, so a transit node does not have the knowledge of the source. 
> From an operational point of  view, what we do today is that we 
> collect netflow statistics on core routers, we project the label stack 
> onto the routing with an external tool to get the Ingress to Egress 
> LDP traffic including the mapping of the flows on the links.
>
> Now for RSVP, we do have such statistics as the LSP is P2P and has 
> states on every node.
>
> Robert mentioned correctly that SR-TE (especially with MPLS dataplane) 
> has limited TE features (we cannot mimic all what RSVP does in SRTE 
> without adding too much complexity).
>
> Thus, is it a problem (transit node stats) worth to be solved ? If 
> yes, where do we accept to put the complexity ? For a stats issue I 
> would rather prefer to move the complexity to an external tool that 
> can do correlations or whatever operations rather than getting it in 
> the forwarding plane…
>
> IMO, that’s a “nice to have” problem to solve getting that we do not 
> have this for LDP and we know the limitations of SR-TE MPLS.
>
> However, Ingress stats per SRTE LSP are for sure mandatory to get !
>
> The main drawback I see with the proposed solution is that it mimics 
> what Entropy label does with a solution which is similar and at the 
> same time cannot replace entropy label as the provided entropy is far 
> from being sufficient (this is not the goal I know, but I was looking 
> for potential use case optimizations). So in a network running entropy 
> label and this mechanism, a router will need to find the ELI/EL and 
> hash, then find another special label to build the stats (maybe 
> tomorrow there will be a third one to look at and a fourth one…). That 
> starts to be a big overhead for the forwarding plane.
>
> Brgds,
>
> Stephane
>
> *From:*mpls [mailto:[e-mail]] *On Behalf Of *Robert Raszuk
> *Sent:* Thursday, November 16, 2017 16:23
> *To:* Alexander Vainshtein
> *Cc:* spring; Clarence Filsfils; mpls; Michael Gorokhovsky; 
> [e-mail]; 
> draft-hegde-spring-traffic-accounting-for-sr-paths; Zafar Ali (zali)
> *Subject:* Re: [mpls] [spring] Special purpose labels in 
> draft-hegde-spring-traffic-accounting-for-sr-paths
>
> Folks,
>
> This thread started and the requirements reported clearly stated that 
> all what we need is the ability to account per path traffic on egress 
> nodes.
>
> Now out of the sudden I see requirement popping up to be able to 
> measure per path in transit nodes.
>
> Well you can do it today with SRv6 if your hardware allows or you can 
> do it with RSVP-TE.
>
> SR-MPLS is replacing LDP and adds ability for limited TE. But SR-MPLS 
> never intended to become connection oriented protocol nor architecture.
>
> So I recommend we take a step back here. Or if you like first go and 
> fix basic MPLS LDP LSPs to allow per end to end path accounting in 
> transit nodes then come back here to ask for the same in SR-MPLS. Not 
> the other way around.
>
> Thx
>
> r.
>
> On Nov 16, 2017 16:12, "Alexander Vainshtein" 
> <[e-mail] 
> <mailto:[e-mail]>> wrote:
>
> Greg,
>
> I concur with your position: let’s first  of all agree that ability to 
> measure traffic carried by an SR-TE LSP in a specific transit node is 
> a require OAM function for SR.
>
> I have looked up the SR OAM Use Cases 
> <https://datatracker.ietf.org/doc/draft-ietf-spring-oam-usecase/?include_text=1> 
> draft, and I did not find any relevant use cases there.
>
> The only time measurements are mentioned is a reference to an expired 
> implementation report 
> <https://tools.ietf.org/html/draft-leipnitz-spring-pms-implementation-report-00> 
> draft discussing delay measurements.  Since delay measurements are in 
> any case based on synthetic traffic, and are always end-to-end 
> (one-way or two-way), this reference is not relevant, IMHO, for this 
> discussion.
>
> I have added the authors of the SR OAM Use Cases draft to tis thread.
>
> Regards,
>
> Sasha
>
> Office: [phone] <tel:+972%203-926-6302>
>
> Cell: [phone] <tel:+972%2054-926-6302>
>
> Email: [e-mail] 
> <mailto:[e-mail]>
>
> *From:*mpls [mailto:[e-mail] 
> <mailto:[e-mail]>] *On Behalf Of *Greg Mirsky
> *Sent:* Thursday, November 16, 2017 4:28 AM
> *To:* Xuxiaohu <[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]>>
> *Subject:* Re: [mpls] [spring] Special purpose labels in 
> draft-hegde-spring-traffic-accounting-for-sr-paths
>
> Dear All,
>
> I cannot imagine that operators will agree to deploy network that 
> lacks critical OAM tools to monitor performance and troubleshoot the 
> network. True, some will brave the challenge and be the early adopters 
> but even they will likely request that the OAM toolbox be sufficient 
> to support their operational needs. I see that this work clearly 
> describes the problem and why ability to quantify the flow behavior at 
> internal nodes is important for efficient network operation. First 
> let's discuss whether the case and requirement towards OAM is real and 
> valid. Then we can continue to discussion of what measurement method 
> to use.
>
> Regards,
>
> Greg
>
> On Thu, Nov 16, 2017 at 10:05 AM, Xuxiaohu <[e-mail] 
> <mailto:[e-mail]>> wrote:
>
>     Concur. Although it has some values, it's not cost-efficient from
>     my point of view. Network simplicity should be the first priority
>     object. Hence we would have to make some compromise.
>
>     Best regards,
>     Xiaohu
>
>
>     ------------------------------------------------------------------------
>
>     徐小虎 Xuxiaohu
>     M：[phone] <tel:[phone]>
>     E：[e-mail] <mailto:[e-mail]>
>     产品与解决方案-网络战略与业务发展部
>     Products & Solutions-Network Strategy & Business Development Dept
>
>     *发件人：***Zafar Ali (zali)
>
>     *收件人：***Greg Mirsky<[e-mail]
>     <mailto:[e-mail]>>;draft-hegde-spring-traffic-accounting-for-sr-paths<[e-mail]
>     <mailto:[e-mail]>>;mpls<[e-mail]
>     <mailto:[e-mail]>>;spring<[e-mail]
>     <mailto:[e-mail]>>
>
>     *主**题：***Re: [mpls] [spring] Special purpose labels in
>     draft-hegde-spring-traffic-accounting-for-sr-paths
>
>     *时间：***2017-11-16 02:24:10
>
>     Hi,
>
>     This draft breaks the SR architecture. I am quoting a snippet from
>     abstract of SR Architecture document
>     https://tools.ietf.org/html/draft-ietf-spring-segment-routing-13,
>     which states:
>
>     “SR allows to enforce a flow through any topological path while
>     maintaining per-flow state only at the ingress nodes to the SR
>     domain.”
>
>     In addition to creating states at transit and egress nodes, the
>     procedure also affects the data plane and makes it unscalable. It
>     also makes controller job much harder and error prune. In summary,
>     I find the procedure very complex and unscalable.
>
>     Thanks
>
>     Regards … Zafar
>
>     *From: *spring <[e-mail]
>     <mailto:[e-mail]>> on behalf of Greg Mirsky
>     <[e-mail] <mailto:[e-mail]>>
>     *Date: *Wednesday, November 15, 2017 at 11:10 AM
>     *To: *"[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]>" <[e-mail] <mailto:[e-mail]>>
>     *Subject: *[spring] Special purpose labels in
>     draft-hegde-spring-traffic-accounting-for-sr-paths
>
>     Hi Shraddha,
>
>     thank you for very well written and thought through draft. I have
>     these questions I'd like to discuss:
>
>       * Have you thought of using not one special purpose label for
>         both SR Path Identifier and SR Path Identifier+Source SID
>         cases but request two special purpose labels, one for each
>         case. Then the SR Path Identifier would not have to lose the
>         bit for C flag.
>       * And how you envision to collect the counters along the path?
>         Of course, a Controller may query LSR for all counters or
>         counters for the particular flow (SR Path Identifier+Source
>         SID). But in addition I'd propose to use in-band mechanism,
>         perhaps another special purpose label, to trigger the LSR to
>         send counters of the same flow with the timestamp out-band to
>         the predefined Collector.
>       * And the last, have you considered ability to flush counters
>         per flow. In Scalability Considerations you've stated that
>         counters are maintained as long as collection of statistics is
>         enabled. If that is on the node scope, you may have to turn
>         off/on the collection to flush off some old counters. I think
>         that finer granularity, per flow granularity would be useful
>         for operators. Again, perhaps the flow itself may be used to
>         signal the end of the measurement and trigger release of counters.
>
>     Regards,
>
>     Greg
>
>
> ___________________________________________________________________________
>
> This e-mail message is intended for the recipient only and contains 
> information which is
> CONFIDENTIAL and which may be proprietary to ECI Telecom. If you have 
> received this
> transmission in error, please inform us by e-mail, phone or fax, and 
> then delete the original
> and all copies thereof.
> ___________________________________________________________________________
>
>
> _______________________________________________
> mpls mailing list
> [e-mail] <mailto:[e-mail]>
> https://www.ietf.org/mailman/listinfo/mpls
>
> _________________________________________________________________________________________________________________________
>
> Ce message et ses pieces jointes peuvent contenir des informations confidentielles ou privilegiees et ne doivent donc
> pas etre diffuses, exploites ou copies sans autorisation. Si vous avez recu ce message par erreur, veuillez le signaler
> a l'expediteur et le detruire ainsi que les pieces jointes. Les messages electroniques etant susceptibles d'alteration,
> Orange decline toute responsabilite si ce message a ete altere, deforme ou falsifie. Merci.
>
> This message and its attachments may contain confidential or privileged information that may be protected by law;
> they should not be distributed, used or copied without authorisation.
> If you have received this email in error, please notify the sender and delete this message and its attachments.
> As emails may be altered, Orange is not liable for messages that have been modified, changed or falsified.
> Thank you.
>
>
> _______________________________________________
> mpls mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/mpls