Re: [Bier] proposed BIER charter

[Antoni Przygienda <[e-mail]>]

<hat type="wg-chair">

Stewart, I concur with Alia that a new data plane concept like this one is a thing seldom introduced and frail with surprising risks (IPv6 = ~15 years from idea to reasonable scale of deployment ? MPLS = ~ 10 years ? ;-)

IMO, it is far less risky to give this thing a vibrant WG where people know that they have to push for serious implementations (and prove them viable before standards track with resulting huge investment in silicon by many companies will be taken) rather than rush it without adequate experience and end up with a massive egg on our collective faces.  The path to standards is fair (FYI, it was initially much stricter) and clearly spelled out me thinks. I'm waiting for people getting excited by the fame attached to a first Linux kernel implementation going @ 1GB bitrate (I'm too smart to talk about packet size or bitstring lengths here ;-)

Pen-ultimately, I would like to see precise use-cases for the low link-speeds and edge to consider this further. A 'may be useful' does not justify major surgery on charter IMO.

Ultimately, early vendor support & strong customer response in the market place is the best pressure point for IETF to move such work into Standards Track lest they end up looking like Johnie-come-laties. We ain't there yet.

</hat>

--- tony

From: BIER [mailto:[e-mail]] On Behalf Of Stewart Bryant
Sent: Thursday, February 12, 2015 5:45 AM
To: Alia Atlas; [e-mail]
Cc: Adrian Farrel; [e-mail]; Alvaro Retana
Subject: Re: [Bier] proposed BIER charter

Alia

I think that you are crossing far too many bridges before you get to them
in terms of you expectation of where BIER will be deployed and how
it will be implemented.

The charter should not include such specific assumptions, particularly
given the fast moving changes that are happening in our industry
especially in the area of packet forwarding design.

I have not seen such assumptions in other charters and I don't think
that they have a place in this one.

In terms of choice of document stream (experimental vs PS),
that is a decision that can be taken at the time of publication
when more information will be available in terms of potential
market take-up, expected/actual deployment scenarios,
breadth of implementation and implementation experience.

Whilst this technology may be deployed in the Internet core,
it my also be deployed at the edge where a much lower threshold
is applicable.

I fear that by introducing these caveats the IETF is entrenching the
perception that it slows technology down rather than embracing
and encouraging innovation.

I would therefore suggest removing the text on implementation
styles and document stream constraints.

- Stewart

On 11/02/2015 20:44, Alia Atlas wrote:
I have been working on getting a charter together for BIER with the intent of pushing for it to be chartered before the Dallas IETF.  This has not yet gone through IESG review and it may have some aspects updated.

Please send comments here.

The charter can be found at http://datatracker.ietf.org/doc/charter-ietf-bier/
and is included below as well.


WG Chairs:
  Greg Shepherd  <[e-mail]<mailto:[e-mail]>>
  Tony Przygienda <[e-mail]<mailto:[e-mail]>>


In conventional IP multicast forwarding, the packets of a given
multicast "flow" are forwarded along a tree that has been constructed
for the specific purpose of carrying that flow.  This requires transit
nodes to maintain state on a per-flow basis, and requires the transit
nodes to participate in multicast-specific tree building protocols.
The flow to which a packet belongs is determined by its IP source and
destination address fields.

BIER (Bit Index Explicit Replication) is an alternative method of
multicast forwarding.  It does not require any multicast-specific
trees, and hence does not require any multicast-specific tree building
protocols.  Within a given "BIER domain", an ingress node encapsulates
a multicast data packet in a "BIER header".  The BIER header
identifies the packet's egress nodes in that domain.  Each possible
egress node is represented by a a single bit within a bitstring; to
send a packet to a particular set of egress nodes, the ingress node
sets the bits for each of those egress nodes, and clears the other
bits in the bistring.  Each packet can then be forwarded along the
unicast shortest path tree from the ingress node to the egress nodes.
Thus there are no per-flow forwarding entries.

Due to the particular sensitivity of adding new significant
functionality into the data-plane at high link speeds, the BIER work
will progress as Experimental.  As described in item (9) below, the
work may become Standards Track once there is sufficient experience
with the benefits and downsides of the technology.

BIER is initially chartered to do experimental work on this new
multicast forwarding mechanism as follows:

   1) BIER architecture: The WG will publish an architecture, based
   upon draft-wijnands-bier-architecture-04.  It will include the
   normative algorithm for how BIER packet forwarding is done.  It
   will specify the information that is required by a BIER header to
   support BIER forwarding.

   2) BIER encapsulation: The working group should assume that the
   technology will need to be embedded in the data plane and operate
   at the highest packet line speeds.  The WG will publish a document
   defining an MPLS-based encapsulation based upon
   draft-wijnands-mpls-bier-encapsulation-02. Due to the critical need
   to have a high-quality and stable RFC for a new data-plane
   encapsulation, the MPLS-based encapsulation draft shall wait after
   WGLC and not progress to IETF Last Call until there are two
   independent interoperable implementations.

   As a secondary focus, the WG may also work on one non-MPLS
   data-plane encapsulation.  This draft also shall wait after WGLC
   and not progress to IETF Last Call until there are two independent
   interoperable implementations.  This draft must focus on and
   include the following details:

       a) What is the applicability of the encapsulation and for which
       use-cases is this encapsulation required?

       b) Does this proposed encapsulation imply any changes to the
       MPLS-based encapsulation?

       c) What design choices have been made for the encapsulation
       type and the included fields.

       d) The proposed encapsulation with considerations given to at
       least OAM, Class of Service, security, fragmentation, TTL.

   3) Transition Mechanisms: The WG will describe how BIER can be
   partially deployed and still provide useful functionality.  A
   minimum of the necessary mechanisms to support incremental
   deployment and/or managing different BIER mask-length compatibility
   may be defined.  Each such mechanism must include an applicability
   statement to differentiate its necessity from other proposed
   mechanisms.

   4) Applicability Statements: The WG will work on a document
   describing how BIER can be applied to multicast L3VPN and to EVPN.
   This draft will describe what mechanism is used to communicate the
   group membership between the ingress router and the egress routers,
   what scalability considerations may arise, and any deployment
   considerations.

   5) Use Case: The WG may produce one use-case document that clearly
   articulates the potential benefits of BIER for different use-cases.
   This would be based upon draft-kumar-bier-use-cases-01.

   6) OAM: The WG will describe how OAM will work in a BIER domain and
   what simplifications BIER offers for managing the multicast
   traffic.  A strong preference will be given to extensions to
   existing protocols.

   7) Management models: The WG may work on YANG models and, if needed,
   MIB modules to support common manageability.

   8) IGP extensions.  When a BIER domain falls within a "link state IGP""
   network, the information needed to set up the BIER forwarding tables
   (e.g., the mapping between a given bit position and a given egress
   router) may be carried in the link state advertisements of the IGP.  The
   link state advertisments may also carry other information related to
   forwarding (e.g., the IGP may support multiple topologies, in which case
   it may be necessary to advertise which topologies are to be used for BIER
   forwarding).  Any necessary extensions to the IGP will be specified by
   the WG, in cooperation with the ISIS and OSPF WGs.

   9) Deployment Experience: Once there is deployment experience, the
   WG will produce a document describing the benefits, problems, and
   trade-offs for using BIER instead of traditional multicast
   forwarding mechanisms.  Ideally, this should also contain an
   analysis of the impact and benefit of the new BIER data-plane to
   the overall Internet architecture.  This document is intended to be
   used to evaluate whether to recharter BIER to produce Standards
   Track RFCs.

The BIER working group will coordinate with several different working
groups and must include the relevant other working groups during
working group last call on the relevant drafts.  BIER will coordinate
with MPLS on the MPLS-based encapsulation and associated MPLS-based
OAM mechanisms.  BIER will coordinate with ISIS and OSPF on extensions
to flood BIER-related information.  BIER will coordinate with BESS and
IDR on the applicability of existing BGP-based mechanisms for
providing multicast group membership information.

Regards,
Alia




_______________________________________________

BIER mailing list

[e-mail]<mailto:[e-mail]>

https://www.ietf.org/mailman/listinfo/bier




--

For corporate legal information go to:



http://www.cisco.com/web/about/doing_business/legal/cri/index.html