Message Multiplexing (memux) Charter

Name | Area | Chair | Directors | Mailing List | Description | Goals & Milestones | Background


Working Group Name

Message Multiplexing (memux)

IETF Area

Transport Area

Chair

Transport Area Directors

Responsible Area Director

Vern Paxson

Mailing List

The <ietf-memux@w3.org> (NOT EXTANT YET; use ietf-http-ng@w3.org until then) mailing list and archives are available for discussions of MEMUX. Postings to this mailing list from non-subscribers are moderated in order to avoid spam, everything else will be passed as is. See the Mailing list administrativia for details.

Description of Working Group

The goal of this working group is to develop a multiplexing protocol to provide a certain set of services, articulated below.  MEMUX is not an attempt to solve all of the world's problem, nor even all the ones articulated at the RUTS BOF at IETF-43 (http://www.ietf.org/proceedings/98dec/43rd-ietf-98dec-142.html).  The MEMUX protocol is to deliver multiplexed bidirectional reliable ordered message streams over a bidirectional reliable ordered byte stream protocol (such as TCP).  Message streams may be initiated by either side, once the underlying byte stream connection is established.  The length of a message is unrestricted (e.g., not bounded by layer 2 or 3 packet sizes), and the payload of a message is also unrestricted; such a message can be used directly, e.g., as a request or a response in an application-level request/response protocol.  Within each message stream, the messages are delivered reliably and in order (as are bytes in TCP).  Each message may be passed as a series of chunks, so that the multiplexing does not introduce unnecessary synchronization between streams.  The MEMUX protocol will be layered on top of bidirectional reliable ordered byte stream protocols (such as, but not limited to, TCP), and multiplex many message streams over a single byte stream connection.  It should be possible for there to be multiple message chunks in one IP packet.  The MEMUX protocol will be lightweight in these two ways: (1) its overhead, in bytes on the wire, will be low, and (2) opening and closing new message streams, once the byte stream connection is established, will take few bytes and impose no round-trip delays.  The value of the MEMUX protocol is twofold: (1) it provides a commonly useful service abstraction (bidirectional reliable ordered arbitrary-sized message stream), and (2) the multiplexing achieves the same results as state sharing between parallel TCP streams (which is not widely available today).  The second value may cease to be unique in the future (when TCP and/or replacements that effectively share state between parallel connections become widely available), but having built other protocols and implementations on top of the service provided by MEMUX enables a smooth transition to a MEMUX-- that delivers the same service while doing no multiplexing of its own.

MEMUX will be designed with security in mind, even though MEMUX's job is not to add security enhancements to protocol stacks.  That is, the "heavy lifting" of implementing security enhancements such as authentication, integrity, privacy, authorization, etc. are to be done in other protocol layers (above and/or below MEMUX); examples include: TLS below, something GSS-based above.  It should be possible to include MEMUX in a protocol stack that does have real security without MEMUX losing the security gained by the other layers.  The WG will consider the issues of security problems introduced by the MEMUX layer itself.  Firewall issues will be considered.

Due consideration will be given to the concerns of the IPv4->v6 transition and the use of NATs.

Some details of requirements remain to be worked out, depending on the interests of, and needs represented by, the participants.  One such issue is the desired kind(s) and degree(s) of insluation, or even control of interaction, between message streams multiplexed over the same byte stream.  Another is the desired kind(s) of endpoint identifiers.  Another is whether MEMUX, perhaps with a slight adaptation layer, is to be usable by MEMUX-unaware applications through an interface designed to present TCP's service (e.g., sockets); this in turn raises a question (in some people's minds) of whether both peers in such a scenario should be required to have MEMUX in their stacks.

Interest in a multiplexing protocol appeared at IETF-43 in the HTTPNG, RUTS, SIGTRAN, MEGACO, and AAA meetings.  The HTTP-NG proponents submitted a draft multiplexing protocol (draft-gettys-webmux-00.txt) on 1 August 1998 as part of the HTTP-NG suite; that protocol could serve as MEMUX, depending on how the open requirements questions are settled.

Deliverables

  1. An Informational RFC on the goals of the MEMUX protocol.
  2. Standards-track specification of the MEMUX protocol.

Out of Scope

  1. Inventing new byte stream protocols; MEMUX is to use existing (e.g., TCP) or new ones.
  2. Underlying connection agility; MEMUX will transport a given message stream over exactly one byte stream.
  3. Datagrams; MEMUX provides message stream connections.
  4. Control channel vs. data channel separation; that distinction is to be made at a higher layer, which will use MEMUX's service for whichever channel(s) it chooses.

Goals and Milestones