Re: [Gen-art] Genart early review of draft-ietf-opsawg-ipfix-bgp-community-04

["Dongjie (Jimmy)" <[e-mail]>]

Hi Joel, 

Thanks a lot for your comments and discussion. 

There may be some misunderstanding about the description of the targeted use cases of this draft, in next revision more texts will be added to clarify this. We will also try to resolve the other comments you raised. 

Just some clarification about BGP communities: 

Except for a few well-known BGP communities, the meaning and behavior of most BGP communities are not standardized, which gives operators the flexibility to use them for many different purposes. RFC4384 describes one use case of BGP communities, RFC8195 gives some examples about the application of BGP large communities, and there are also many use cases which are not documented in RFCs. 

One typical usage of BGP communities is to represent geographical or customer information, this is also described in RFC 8195 for the newly defined large communities. In such use case, it is useful for operators to collect the traffic statistics based on BGP communities, as this provides the aggregation of traffic flows with reasonable granularity.

Best regards,
Jie

> -----Original Message-----
> From: Joel Halpern Direct [mailto:[e-mail]]
> Sent: Wednesday, February 28, 2018 11:19 PM
> To: li zhenqiang <[e-mail]>; Dongjie (Jimmy)
> <[e-mail]>; [e-mail]
> Cc: [e-mail]; opsawg
> <[e-mail]>
> Subject: Re: Genart early review of draft-ietf-opsawg-ipfix-bgp-community-04
> 
> I am having trouble reconciling two of your comments.
> In you rlast email you said that this is for "planed communities represent the
> groups of customers peers an geographical and topological related
> information".  Planned communities is presumably a new behavior, not
> existing behavior.
> 
> In this email you say that these are "already defined BGP communities".
> 
> You reference RFC 4384, which talks about several kinds of communities.
> maybe you intend the regional or national communities mentioned as being
> possible, but not defined, in that document.  This document's descriptions do
> not align well enough with RFC 4384 for me to say.
> 
> Let's be clear.  The working group asked for an early review.  The WG now
> has my review, indicating that this document is unclear in multiple regards and
> could use improvement.
> 
> It is now up to the WG and the chairs.
> Yours,
> Joel
> 
> On 2/28/18 6:22 AM, li zhenqiang wrote:
> > Hi Joel,
> >
> > This is not for one operator, instead it is a common practice. Please
> > refer to RFC4384 and comments from Thomas who are from Swisscom.
> >
> > One clarification for this doc is it is not to introduce any new BGP
> > communities but to report the already defined BGP communities related
> > to a traffic flow through IPFIX, thus the IPFIX collector can analyze
> > the traffic in BGP community granularity without running BGP protocol.
> >
> > BGP community is a transitive attibute, thus the exporter can report
> > all the communities carried in the matching route entry, unless some
> > BGP communities are filtered by some routers.
> >
> > Sure I can add some text in the doc to say the proper processing of
> > the exporter, something like what I said in the previous mail, do you
> > think it is ok and enough?
> >   When the exporter, i.e. router, receives the templete to report the
> > communities, the exporter gets the information through BGP lookup
> > using the corresponding source or destination IP of a traffic flow.
> >
> > Thank you for your comments.
> >
> > Best Regards,
> > Zhenqiang Li
> > ----------------------------------------------------------------------
> > --
> > [e-mail]
> >
> >     *From:* Joel M. Halpern <mailto:[e-mail]>
> >     *Date:* 2018-02-28 10:13
> >     *To:* li zhenqiang <mailto:[e-mail]>; Dongjie
> >     (Jimmy) <mailto:[e-mail]>; [e-mail]
> >     <mailto:[e-mail]>
> >     *CC:* [e-mail]
> >     <mailto:[e-mail]>; opsawg
> >     <mailto:[e-mail]>
> >     *Subject:* Re: Genart early review of
> >     draft-ietf-opsawg-ipfix-bgp-community-04
> >     Is this for one operator (still important, but not necessarily for
> >     standardization) or are there several operators who have expressed
> >     interest in this?
> >     Yes, we do proactive standards.  But the IDR group, for example,
> tends
> >     to be very careful to see if interest is reflected in implementation.
> >     In this case, given that what is proposed is a completely different use
> >     of the BGP communities, I think at least more clarity that this is only
> >     expected to be used for communities that match the purpose, and of
> how
> >     and why the vendors would implement the router-side logic.
> >     To get back to the points I made in the review:
> >     1) The document needs to be much clearer that it is about new
> >     communities whcih are expected to be defined for this use.  It needs
> to
> >     be clear if this is expected to be applied to communities put on by
> >     other AS, or only to communities provided by routers of the collecting
> >     AS.  The later leads to understandable configuration.  The former
> leads
> >     to questions about hos the meaning will be known.
> >     2) The document needs to be clear and explicit about what processing it
> >     is expecting the router to provide, and how much configuration is
> >     needed
> >     to get the right things to happen.
> >     Yours,
> >     Joel
> >     On 2/27/18 8:54 PM, li zhenqiang wrote:
> >      > Hi Joel,
> >      >
> >      > This is Zhenqiang Li from China Mobile. The purpose of this doc
> >     is not
> >      > to report the well-known communities, but the operator planed
> >      > communities represent the groups of the customers, peers,
> >      > the geographical and topological related information as stated
> in
> >      > RFC4384, which is a common practice and also used in our field
> >     network.
> >      >
> >      > When the exporter, i.e. router, receives the templete to report the
> >      > communities, the exporter gets the information through BGP lookup
> >     using
> >      > the corresponding source or destination IP of a traffic flow. The
> >      > procedure for the exporter to get the community informaiton of a
> >     traffic
> >      > flow is the same as it gets the AS information.
> >      >
> >      > Best Regards,
> >      > Zhenqiang Li
> >      >
> >     ------------------------------------------------------------------------
> >      > [e-mail]
> >      >
> >      >     *From:* Joel M. Halpern <mailto:[e-mail]>
> >      >     *Date:* 2018-02-12 00:37
> >      >     *To:* Dongjie (Jimmy) <mailto:[e-mail]>;
> >     [e-mail]
> >      >     <mailto:[e-mail]>
> >      >     *CC:* [e-mail]
> >      >     <mailto:[e-mail]>;
> >      >     [e-mail] <mailto:[e-mail]>
> >      >     *Subject:* Re: Genart early review of
> >      >     draft-ietf-opsawg-ipfix-bgp-community-04
> >      >     This was a requested early review.  You folks can do as you
> >     deem best.
> >      >      From where I sit, it seems odd.  Most well-known
> communities
> >     do not
> >      >     fit
> >      >     the pattern of representing groups of sources or groups of
> >     destinations.
> >      >     I presume the intent here is for this to be useful in some AS
> >     other
> >      >     than
> >      >     the one originating the communities.  Which makes it even
> >     harder to see
> >      >     when it would apply.
> >      >     I presume this is driven by having found that it would have
> >     helped in
> >      >     some real-world situation?
> >      >     I think the document would be helped by a clearer description
> of
> >      >     when it
> >      >     applies and what behavior is expected of the router (not just
> >     "the same
> >      >     as that over there.")
> >      >     Yours,
> >      >     Joel
> >      >     On 2/11/18 1:32 AM, Dongjie (Jimmy) wrote:
> >      >      > Hi Joel,
> >      >      >
> >      >      > Thanks for your review comments. Please see my replies
> inline:
> >      >      >
> >      >      >> -----Original Message-----
> >      >      >> From: Joel Halpern [mailto:[e-mail]]
> >      >      >> Sent: Saturday, February 10, 2018 1:27 AM
> >      >      >> To: [e-mail]
> >      >      >> Cc: [e-mail];
> >      >     [e-mail]
> >      >      >> Subject: Genart early review of
> >      >     draft-ietf-opsawg-ipfix-bgp-community-04
> >      >      >>
> >      >      >> Reviewer: Joel Halpern
> >      >      >> Review result: Not Ready
> >      >      >>
> >      >      >> This is an early gen-art review of
> >     draft-ietf-opsawg-ipfix-bgp-04.
> >      >      >>
> >      >      >> The document is clear about what it is trying to do, and
> >      >     readable.  It is not
> >      >      >> clear about how it expects this to actually work.
> >      >      >>
> >      >      >> However, I find the underlying concept confusing.
> >      >      >> 1) BGP Communities may sometimes represent subsets of
> >     traffic.
> >      >     But usually
> >      >      >> they represent tagging intended to influence routing
> which is
> >      >     only indirectly
> >      >      >> related to meaningful subsets of traffic for TE
> >     purposes.  One
> >      >     may be able to
> >      >      >> make an argument that this could better enable
> monitoring the
> >      >     effects of some
> >      >      >> BGP communities.  But the draft does not make that
> argument.
> >      >      >
> >      >      > This depends on how the BGP communities are used by the
> >      >     operators. Except some well-known communities, BGP
> >     communities are
> >      >     used in a customized manner. In some cases, BGP
> communities
> >     indicate
> >      >     the source and destination information of a group of traffic
> >     flows.
> >      >     These are the major case this document is focusing on, as it
> >     would
> >      >     be helpful for operator to collect the traffic statistics
> >     based on
> >      >     BGP communities. Using BGP communities to influence
> routing is
> >      >     another popular use case. In that case, it may also be helpful
> to
> >      >     collect traffic statistic information related to the BGP
> >      >     communities, while the purpose may not be just for TE.
> >      >      >
> >      >      > 2) It is
> >      >      >> unclear what this actually expects the router to do in
> >      >     generating this
> >      >      >> information.
> >      >      >> Reading between the lines, it seems that what is desired
> >     is for
> >      >     the router
> >      >      >> control process to go through the IPFIX collected
> information
> >      >     before it is
> >      >      >> exported, and add BGP community tags to the export
> >     information.
> >      >      >> (Generating such information directly from the forwarding
> >     plane
> >      >     would place
> >      >      >> significant load on the forwarding representation and
> >      >     processing, and on the
> >      >      >> control logic to generate FIB information.)  Given that
> >     off-line
> >      >     BGP information
> >      >      >> collection is a common practice, and that such
> information is
> >      >     common across
> >      >      >> the AS, it would actually seem simpler to perform such
> >      >     processing and
> >      >      >> aggregation offline rather than in the router.
> >      >      >
> >      >      > The behavior of a router would be similar to its behavior
> with
> >      >     the existing BGP relevant IEs, e.g. bgpSourceAsNumber,
> >      >     bgpDestinationAsNumber, bgpNextHopIPv4Address, etc.
> Basically
> >     this
> >      >     is the aggregated traffic information collection model, in
> >     which the
> >      >     router aggregates the collected traffic information based on
> >     the IEs
> >      >     specified in the template, so that it can export much less
> >      >     information to the collector without losing the information the
> >      >     collector really cares about. Exporting aggregated traffic
> >      >     statistics has been widely used in the networks.
> >      >      >
> >      >      > Note that the purpose of this mechanism is to export the
> >      >     aggregated traffic statistics information at the granularity
> >      >     specified by BGP communities, while BMP can used to collect
> the
> >      >     detailed information of BGP RIBs and BGP events, IMO they
> are
> >      >     designed for different purposes. Although it is possible to
> >     export
> >      >     all the non-aggregated traffic information to the collector,
> >     and let
> >      >     the collector to correlate them with the BGP communities,
> >     this can
> >      >     bring heavy burden to both the exporter and the collector.
> >      >      >
> >      >      >>
> >      >      >> If the IDR working group has not been consulted about
> this, I
> >      >     would strongly
> >      >      >> recommend working with them as to whether this is
> actually
> >      >     useful information
> >      >      >> to collect, and how and where to collect it. If the IDR
> >     working
> >      >     group does not
> >      >      >> consider important to work on this, then that gives you
> >     useful
> >      >     information in
> >      >      >> and of itself.
> >      >      >
> >      >      > The IDR WG has been notified about the LC of this
> document, so
> >      >     far there is no objection received from them. We would like to
> >      >     encourage IDR people to review and give feedbacks to help
> improve
> >      >     this document. Whether the new IEs are useful or not should
> be
> >      >     determined in the OPSAWG.
> >      >      >
> >      >      > Best regards,
> >      >      > Jie
> >      >      >
> >      >
> >