HyBi Working Group J. Tamplin Internet-Draft T. Yoshino Intended status: Standards Track Google, Inc. Expires: January 5, 2013 July 4, 2012 A Multiplexing Extension for WebSockets draft-ietf-hybi-websocket-multiplexing-03 Abstract The WebSocket Protocol [RFC6455] requires a new transport connection for every WebSocket connection. This presents a scalability problem when many clients connect to the same server, and is made worse by having multiple clients running in different tabs of the same user agent. This extension provides a way for separate logical WebSocket connections to share an underlying transport connection. Please send feedback to the hybi@ietf.org mailing list. Status of this Memo This Internet-Draft is submitted to IETF in full conformance with the provisions of BCP 78 and BCP 79. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet- Drafts is at http://datatracker.ietf.org/drafts/current. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." This Internet-Draft will expire on January 5, 2013. Copyright Notice Copyright (c) 2012 IETF Trust and the persons identified as the document authors. All rights reserved. This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must Tamplin & Yoshino Expires January 5, 2013 [Page 1] Internet-Draft A Multiplexing Extension for WebSockets July 2012 include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License. Table of Contents 1. Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.1. Physical Connection and Logical Channels . . . . . . . . . 3 2. Conformance Requirements . . . . . . . . . . . . . . . . . . . 4 3. Interaction with other Extensions / Framing Mechanisms . . . . 5 3.1. Choosing the point to apply an extension . . . . . . . . . 5 4. Multiplexed Connections . . . . . . . . . . . . . . . . . . . 7 5. Flow Control . . . . . . . . . . . . . . . . . . . . . . . . . 8 5.1. New Channel Slot . . . . . . . . . . . . . . . . . . . . . 8 5.2. Send Quota . . . . . . . . . . . . . . . . . . . . . . . . 8 6. Framing . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 7. Multiplex Control Blocks . . . . . . . . . . . . . . . . . . . 12 7.1. AddChannelRequest . . . . . . . . . . . . . . . . . . . . 12 7.2. AddChannelResponse . . . . . . . . . . . . . . . . . . . . 15 7.3. FlowControl . . . . . . . . . . . . . . . . . . . . . . . 16 7.4. DropChannel . . . . . . . . . . . . . . . . . . . . . . . 17 7.5. NewChannelSlot . . . . . . . . . . . . . . . . . . . . . . 19 8. Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 9. Client Behavior . . . . . . . . . . . . . . . . . . . . . . . 22 10. Buffering . . . . . . . . . . . . . . . . . . . . . . . . . . 23 11. Fairness . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 12. Proxies . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 13. Nesting . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 14. Timeout . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 15. Close the Logical Channel . . . . . . . . . . . . . . . . . . 28 16. Fail the Logical Channel . . . . . . . . . . . . . . . . . . . 29 17. Fail the Physical Connection . . . . . . . . . . . . . . . . . 30 18. Operations and Events on Multiplexed Connection . . . . . . . 31 19. Security Considerations . . . . . . . . . . . . . . . . . . . 32 20. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 33 21. Normative References . . . . . . . . . . . . . . . . . . . . . 34 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 35 Tamplin & Yoshino Expires January 5, 2013 [Page 2] Internet-Draft A Multiplexing Extension for WebSockets July 2012 1. Overview This document describes a multiplexing extension for the WebSocket Protocol. With this extension, one TCP connection can provide multiple virtual WebSocket connections by encapsulating frames taged with a channel ID. A client that supports this extension will advertise support for it in the client's opening handshake using the "Sec-WebSocket-Extensions" header. If the server supports this extension and supports parameters compatible with the client's request, it accepts the use of this extension by the "Sec-WebSocket-Extensions" header in the server's opening handshake. 1.1. Physical Connection and Logical Channels Under use of this extension, one transport connection is shared by multiple application-level instances. The WebSocket connection which lies directly on the TCP connection and negotiated this multiplexing extension is called "physical connection". Virtual WebSocket connections established for each application-level instance are called "mutiplexed connections". Data channels virtually established by ID tagging are called "logical channels". Logical channels with non-zero ID exchange data for multiplexed connections. The logical channel with ID of 0 exchanges multiplex control information. Data for different logical channels are distinguished by the channel ID placed at the head of the message that encapsulates the original frame of a multiplex connection. Tamplin & Yoshino Expires January 5, 2013 [Page 3] Internet-Draft A Multiplexing Extension for WebSockets July 2012 2. Conformance Requirements All diagrams, examples, and notes in this specification are non- normative, as are all sections explicitly marked non-normative. Everything else in this specification is normative. The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC2119 [RFC2119]. Requirements phrased in the imperative as part of algorithms (such as "strip any leading space characters" or "return false and abort these steps") are to be interpreted with the meaning of the key word ("must", "should", "may", etc) used in introducing the algorithm. Conformance requirements phrased as algorithms or specific steps MAY be implemented in any manner, so long as the end result is equivalent. (In particular, the algorithms defined in this specification are intended to be easy to follow, and not intended to be performant.) Tamplin & Yoshino Expires January 5, 2013 [Page 4] Internet-Draft A Multiplexing Extension for WebSockets July 2012 3. Interaction with other Extensions / Framing Mechanisms If any extension (e.g. compression) is placed before this extension in the "Sec-WebSocket-Extensions" header of the physical connection, that extension is applied to multiplexed connections unless otherwise noted in the extension's spec. If any extension is placed after this extension in the "Sec-WebSocket-Extensions" header of the physical connection, that extension is applied to frames after multiplexing on the sender side, and before demultiplexing on the receiver side unless otherwise noted in the extension's spec. A client MAY request such an extension for both the physical connection and multiplexed connections by placing extension entries before and after this multiplexing extension. In this case, the server SHOULD reject at least either of them if it's useless to apply the same extension twice. For example, if we have a compression extension called foo-compress, the client sends Sec-WebSocket-Extensions: foo-compress, mux, foo-compress in the client's opening handshake of the physical connection to request use of the compression for both physical and multiplexed connections. Then, the server would send back Sec-WebSocket-Extensions: mux, foo-compress to apply compression after multiplexing, or Sec-WebSocket-Extensions: foo-compress, mux to apply compression to multiplexed connections. 3.1. Choosing the point to apply an extension Where to apply a compression extension makes difference to resource consumption and flexibility. Compression algorithms often use some memory to keep its context. Some of compression extensions may keep using the same context for all the messages on the same connection. If such an extension is applied to the physical connection, intermediaries that want to demultiplex or multiplex the connection need to decompress (before demultiplexing) and recompress (before multiplexing again) all the frames. Tamplin & Yoshino Expires January 5, 2013 [Page 5] Internet-Draft A Multiplexing Extension for WebSockets July 2012 If such an extension is applied to each multiplexed connection, we can control to which channel we apply the compression, so we can avoid applying compression to channels transferring incompressible data. Intermediaries that want to demultiplex can forward payload leaving it untouched. However, compressing each multiplexed connection is expensive in terms of memory consumption. Tamplin & Yoshino Expires January 5, 2013 [Page 6] Internet-Draft A Multiplexing Extension for WebSockets July 2012 4. Multiplexed Connections The multiplexing extension maintains separate logical channels, each of which provides fully the logical equivalent of an independent WebSocket connection, including separate handshake headers. If the multiplexing extension is successfully negotiated, one multiplexed connection is automatically established, and the headers on the opening handshake of the physical connection are automatically taken to mean ones for the multiplexed connection. It's called "Implicitly Opened Connection". It's served by the logical channel with channel ID of 1 which is also implicitly opened on completion of the opening handshake. New channels are added by the client issuing the AddChannelRequest multiplex control block (note that only the client may initiate new WebSocket connections), including any handshake headers which do not have the same value as the client's opening handshake of the physical connection. The server's AddChannelResponse likewise includes any handshake headers which are different from the server's opening handshake of the physical connection Channel 0 (control channel) is reserved for multiplex control blocks and does not contain Payload Data from any multiplexed connection. A client which attempts to add a channel to an existing connection that is not accepted by the server SHOULD attempt to open a new underlying connection and open a new WebSocket connection on it. Once the multiplexing extension is negotiated on a connection, all frames of multiplexed connection MUST be prefixed with a channel ID number and encapsulated into wrapping binary messages. The channel ID is assigned by the client on issuing an AddChannelRequest. A receiver MAY process frames for different non-control logical channels in parallel. A receiver MUST process frames for the control channel exclusively. A receiver MUST _Fail the Physical Connection_ if any of these rules are violated by the sender. Tamplin & Yoshino Expires January 5, 2013 [Page 7] Internet-Draft A Multiplexing Extension for WebSockets July 2012 5. Flow Control 5.1. New Channel Slot A client has a pool of slots called "new channel slots". It's initialized to be empty on establishment of the physical connection. A NewChannelSlot multiplex control block sent by the server adds slots to the pool. Each slot has a non-negative integer value called "initial send quota". Its function is explained in the later subsection. When sending an AddChannelRequest, the client MUST pick the oldest new channel slot from the pool and remove it from the pool. If there are no slots in the pool, the client MUST NOT issue an AddChannelRequest. A server can regulate the rate of AddChannelRequests by not replenishing the pool. 5.2. Send Quota For each logical channel with non-zero ID, server and client are respectively given a non-negative integer value called "send quota". For the logical channel created for the "Implicitly Opened Connection", the client's "send quota" is initialized to 0 on establishment of the physical connection. The server's "send quota" for the channel is initialized on sending its opening handshake for the physical connection. The "quota" extension parameter attached to the extension token for this multiplexing extension in the client's opening handshake for the physical connection specifies the initial value. If the "quota" extension parameter is not specified, the initial value is set to 0. The extension parameter has the initial value on its parameter value side as a non-negative integer in decimal. For a logical channel added by issuing an AddChannelRequest, a client gets "send quota" equal to the "initial send quota" value on the "new channel slot" picked for the AddChannelRequest on sending it. For a logical channel added by accepting an AddChannelRequest, a server gets "send quota" of 0 on sending the corresponding AddChannelResponse. When an endpoint receives a FlowControl for a logical channel, its "send quota" for the channel gets replenished. Tamplin & Yoshino Expires January 5, 2013 [Page 8] Internet-Draft A Multiplexing Extension for WebSockets July 2012 When sending a frame on a logical channel with non-zero ID, the length of the "Payload data" of the frame MUST NOT be greater than the "send quota" of the endpoint for the channel. An endpoint MUST _Fail the Logical Channel_ when it's clear that the other peer violates this. When a frame is sent on a logical channel with non-zero ID, the length of the "Payload data" of the frame is subtracted from the "send quota" of the endpoint for the channel. An endpoint SHOULD NOT delay replenishment of the other peer's "send quota" for a logical channel when it has more room for accepting new data for the channel. Tamplin & Yoshino Expires January 5, 2013 [Page 9] Internet-Draft A Multiplexing Extension for WebSockets July 2012 6. Framing This extension encapsulates each frame of a multiplexed connection into a binary message with Payload Data obtained by concatenating the following data in the order they are listed: 1. The logical channel ID for the multiplexed connection. 2. FIN, RSV1, RSV2, RSV3 and opcode of the original frame. 3. Unmasked "Payload Data" of the original frame. The channel ID is encoded as a variable number of bytes, as follows: 0 1 2 3 4 5 6 7 +-+-------------+ |0|Channel ID(7)| +-+-------------+ 0 1 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 +-+-+---------------------------+ |1|0| Channel ID (14) | +-+-+---------------------------+ 0 1 2 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 +-+-+-+-----------------------------------------+ |1|1|0| Channel ID (21) | +-+-+-+-----------------------------------------+ 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+---------------------------------------------------------+ |1|1|1| Channel ID (29) | +-+-+-+---------------------------------------------------------+ All frames with a non-zero channel ID MUST be delivered to the corresponding multiplexed connection in the order they are received. Control frames with a non-zero channel ID may also trigger additional processing by the multiplexing extension. This extension MAY change the fragmentation of the original message before encapsulation in order to insert multiplex control blocks or adjust the amount of data to flush along with flow control. In doing this, control messages MAY also be fragmented. Control messages multiplexed with fragmentation MUST be delivered to the corresponding multiplexed connection after receiving all fragments and Tamplin & Yoshino Expires January 5, 2013 [Page 10] Internet-Draft A Multiplexing Extension for WebSockets July 2012 defragmenting them. To allow for adjustment of fragmentation, this multiplexing extension MUST NOT be used after any extension that does any of the followings: o Require frame boundary on its output to be preserved. o Use the "Extension data" field or any of the reserved bits on the WebSocket header as per-frame attribute. Intermediaries MAY change the fragmentation of the encapsulating messages. If a binary message doesn't contain a valid channel ID, _Fail the Physical Connection_. The cases where it's considered that a channel ID is invalid are: o The payload of the message doesn't contain a complete channel ID. o No channel has been opened for the channel ID. o The channel has been closed and not reopened. If a binary message with a non-zero channel ID doesn't contain at least one octet after the channel ID, _Fail the Physical Connection_. Tamplin & Yoshino Expires January 5, 2013 [Page 11] Internet-Draft A Multiplexing Extension for WebSockets July 2012 7. Multiplex Control Blocks A binary messages with the channel ID of 0 contain zero or more multiplex control blocks in "Payload data". Putting multiple control blocks into one WebSocket message saves framing overhead. Unless any other negotiated extension defines a meaning for them, endpoints MUST NOT send any data frame with an opcode other than "binary frame". If an endpoint received such frame, it MUST _Fail the Physical Connection_. Control blocks has fields as follows: 0 1 2 3 4 5 6 7 +-----+---------+ | Opc | | +-----+ : | Opc specific : : data : | | +---------------+ Opc A multiplex control opcode as defined in the following subsections. Opc of 5-7 are Reserved for future use (TBD: do we need some support for quiescence?). Opc specific data Data interpreted according to that opcode. Each of the following subsections describes one multiplex control opcode and how to interpret opc specific data for that opcode. If any incomplete multiplex control block is found, the endpoint MUST _Fail the Physical Connection_. 7.1. AddChannelRequest AddChannelRequest is sent only by clients to create the objective logical channel, as if a new connection were received on a separate transport connection, except for the encoding of the headers. Multiplex control opcode of AddChannelRequest is 0. AddChannelRequest has fields as follows: Tamplin & Yoshino Expires January 5, 2013 [Page 12] Internet-Draft A Multiplexing Extension for WebSockets July 2012 0 1 2 3 4 5 6 7 +-+-+-+-+---+---+ |0|0|0|R|Enc|Len| +-+-+-+-+---+---+ |Objective | :channel ID : |(8-32 bit) | +---------------+ |Size of encoded| :handshake : |(8-32 bit) | +---------------+ |Encoded | :handshake : | | +---------------+ R Reserved for future use. Len The size of the size of the encoded handshake field in bytes minus 1. Enc Encoding scheme type: 0 - identity The client's handshake data in the encoded handshake field are sent as-is without any special encoding or compression applied, and constitute the complete set of a Request-Line and headers that would have been sent on opening handshake. 1 - delta-encoded The client's handshake data in the encoded handshake field are delta-encoded, where any header that is not given is assumed to have the same value as that given on the current delta base. The delta base is initially the client's opening handshake of the physical connection but the "Sec-WebSocket-Extensions" in it is replaced with the extension entries after the one for this multiplexing extension. Every time, an AddChannelRequest where Enc field is identity is received, the delta base is updated to the handshake data in the AddChannelRequest. The Tamplin & Yoshino Expires January 5, 2013 [Page 13] Internet-Draft A Multiplexing Extension for WebSockets July 2012 Request-Line MUST be sent regardless if it's the same as one in the delta base or not. A header with an empty value means that the header is not inherited from the delta base. When to send valueless headers, identity encoding MUST be used. 2-3 - reserved Reserved for future use. Objective channel ID The channel ID of the logical channel objective to this operation. Encoding is the same as one used for encapsulation. Size of encoded handshake The size of the encoded handshake field in bytes in network byte order. Encoded handshake The client's opening handshake as defined in Section 4 of RFC 6455 [RFC6455] for the new multiplexed connection. This field is encoded as defined in the description for the Enc field. The complete set of a Request-Line and headers after decoding is treated exactly as if it was sent on a separate connection. If there's already a logical channel with the same channel ID as one specified in this AddChannelRequest, it MUST _Fail the Physical Connection_. Once the server receives an AddChannelRequest, it MUST decide whether to accept or reject the request. To accept the request, it MUST create a new logical channel with the channel ID set to the objective channel ID field of the AddChannelRequest. Channel ID assignment is done by the client. A client MAY use any algorithm to choose channel IDs for new channels. Note that channel ID assignment might be changed by intermediaries, so it's not guaranteed that the value of channel ID is the same on the other peer. The server MUST respond to the request with the AddChannelResponse described later. Different from non-multiplexed WebSocket connection, a client MAY send frames of multiplexed connections except for "Implicitly Opened Connection" before receiving AddChannelResponse as far as there's sufficient send quota. In case the AddChannelRequest fails, those frames are discarded by the other peer. This doesn't mean that users Tamplin & Yoshino Expires January 5, 2013 [Page 14] Internet-Draft A Multiplexing Extension for WebSockets July 2012 of this protocol such as the WebSocket API are required to allow their users to send frames before receiving the server's opening handshake. 7.2. AddChannelResponse AddChannelResponse is sent only by servers in response to the AddChannelRequest. Multiplex control opcode of the AddChannelResponse is 1. AddChannelResponse has fields as follows: 0 1 2 3 4 5 6 7 +-+-+-+-+---+---+ |0|0|1|F|Enc|Len| +-+-+-+-+---+---+ |Objective | :channel ID : |(8-32 bit) | +---------------+ |Size of encoded| :handshake : |(8-32 bit) | +---------------+ |Encoded | :handshake : | | +---------------+ F If F is set, then the server has rejected the AddChannelRequest and this SHOULD be treated exactly the same as if a separate connection was attempted and the connection was closed after receiving the server's handshake. Enc MUST be set to identity in this case. The encoded handshake field MUST contain the response to an HTTP Upgrade request for the request made by the AddChannelRequest, for example: HTTP/1.1 404 Not found 404 message body... If F is not set, then the server has accepted the AddChannelRequest. The encoded handshake field MUST contain the response to an HTTP Upgrade request for the request made by the AddChannelRequest, for example: Tamplin & Yoshino Expires January 5, 2013 [Page 15] Internet-Draft A Multiplexing Extension for WebSockets July 2012 Len The size of the size of encoded handshake field in bytes minus 1. Enc Encoding scheme type the same as defined for the AddChannelRequest opcode (but replacing "AddChannelRequest" with "AddChannelResponse", and "Request-Line" with "Response-Line"). Objective channel ID Same as one in the AddChannelRequest. Size of encoded handshake The size of the following encoded handshake field in bytes in network byte order. Encoded handshake The server's opening handshake as defined in Section 4 of RFC 6455 [RFC6455] for this multiplexed connection. This field is encoded according to Enc. The complete set of a Response-Line and headers after decoding is treated exactly as if it was received in response to a client's opening handshake on a separate connection. If the server's opening handshake is validated, the client MUST take this as _The WebSocket Connection is Established_. 7.3. FlowControl FlowControl is used to replenish the other peer's send quota for the specified logical channel. Multiplex control opcode of FlowControl is 2. FlowControl has fields as follows. Tamplin & Yoshino Expires January 5, 2013 [Page 16] Internet-Draft A Multiplexing Extension for WebSockets July 2012 0 1 2 3 4 5 6 7 +-+-+-+-----+---+ |0|1|0| RSV |Len| +-+-+-+-----+---+ |Objective | :channel ID : |(8-32 bit) | +---------------+ |Replenished | :quota : |(8-32 bit) | +---------------+ RSV Reserved for future use. Len The size of replenished quota field in bytes minus 1. Objective channel ID Same as one in the AddChannelRequest. Replenished quota An unsigned integer in network byte order added to the quota of the number of bytes the receiver can have outstanding towards the sender of the FlowControl message. 7.4. DropChannel DropChannel is used to close a logical channel. Multiplex control opcode of DropChannel is 3. DropChannel has fields as follows: Tamplin & Yoshino Expires January 5, 2013 [Page 17] Internet-Draft A Multiplexing Extension for WebSockets July 2012 0 1 2 3 4 5 6 7 +-+-+-+-+---+---+ |0|1|1|F|RSV|Len| +-+-+-+-+---+---+ |Objective | :channel ID : |(8-32 bit) | +---------------+ |Size of reason | :(8-32 bit) : | | +---------------+ |Reason | : : | | +---------------+ F F is set when this DropChannel is due to multiplexing level error. F is unset when this DropChannel is sent because the multiplexed connection is asked to _Close the WebSocket Connection_, and the reason field MUST be empty for this case. RSV Reserved for future use. Len The size of the size of reason field in bytes minus 1. Objective channel ID Same as one in the AddChannelRequest. Size of reason The size of the reason field in bytes in network byte order. Reason The reason of closure. When an endpoint received DropChannel, the endpoint MUST remove the logical channel and the application instance that used the logical channel MUST treat this as closure of underlying transport. Once a logical channel is removed by the DropChannel request, the channel ID Tamplin & Yoshino Expires January 5, 2013 [Page 18] Internet-Draft A Multiplexing Extension for WebSockets July 2012 of the logical channel becomes available again for new AddChannelRequest commands. 7.5. NewChannelSlot NewChannelSlot is sent only by servers to adds new slots to the client's new channel pool. Multiplex control opcode of NewChannelSlot is 4. NewChannelSlot has fields as follows: 0 1 2 3 4 5 6 7 +-+-+-+-+---+---+ |1|0|0|R|NL |QL | +-+-+-+-+---+---+ |Number of slots| :(8-32 bit) : | | +---------------+ |Initial send | :quota : |(8-32 bit) | +---------------+ R Reserved for future use. NL The size of the number of slots field in bytes minus 1. QL The size of the initial quota field in bytes minus 1. Number of slots The number of slots to add in network byte order. Initial quota The initial quota each of slots added by this NewChannelSlot gets in bytes in network byte order. When a client received a NewChannelSlot, the client MUST add new slots of the specified number. Each of new slots gets the specified Tamplin & Yoshino Expires January 5, 2013 [Page 19] Internet-Draft A Multiplexing Extension for WebSockets July 2012 initial send quota. Tamplin & Yoshino Expires January 5, 2013 [Page 20] Internet-Draft A Multiplexing Extension for WebSockets July 2012 8. Examples _This section is non-normative._ The examples below assume the handshake has already completed and the multiplexing extension was negotiated. Frames of encapsulating messages from client to server MUST be masked. To simplify, the examples below are not masked. 0x82 0x0d 0x01 0x81 "Hello world" This is a non-fragmented text message of "Hello world" on channel 1 encapsulated into a non-fragmented message. 0x02 0x07 0x01 0x81 "Hello" 0x80 0x06 " world" This is a fragmented encapsulating message converying a non- fragmented text frame "Hello world" on channel 1. 0x82 0x07 0x01 0x01 "Hello" 0x82 0x05 0x02 0x81 "bye" 0x82 0x08 0x01 0x80 " world" This example shows how data for two channels are interleaved. There're three non-fragmented encapsulating messages. The first and third one convey each of two frames of a fragmented text message of "Hello world" on channel 1. The second one conveys a non-fragmented text message of "bye" on channel 2. Tamplin & Yoshino Expires January 5, 2013 [Page 21] Internet-Draft A Multiplexing Extension for WebSockets July 2012 9. Client Behavior When a client is asked to _Establish a WebSocket Connection_ by some WebSocket application instance, it MAY choose to reuse an existing WebSocket connection if all of the following are true: o the multiplexing extension was successfully negotiated on that connection o the scheme portions of the URIs match exactly o the host portions of the URIs either match exactly or resolve to the same IP address (TBD: consider DNS rebind attacks) o the port portions of the URIs (either explicit or implied by the scheme) match exactly o the connection has an availablle logical channel ID If the client chooses to reuse an existing multiplexed connection, it sends an AddChannelRequest as described above. If the AddChannelRequest is accepted, WebSocket frames may be sent over that channel as normal. If the server rejects the AddChannel, the client SHOULD attempt to open a new physical WebSocket connection (for example, in a shared hosting environment a server may not be prepared to multiplex connections from different customers despite having a single IP address for them). Tamplin & Yoshino Expires January 5, 2013 [Page 22] Internet-Draft A Multiplexing Extension for WebSockets July 2012 10. Buffering There will be lots of small frames sent in this protocol (particularly replenishing send quotas), so a sender SHOULD attempt to aggregate multiplex control blocks into larger WebSocket frames. For data frames, a sender also SHOULD attempt to aggregate fragments into one packet of the underlying transport. However, care must be taken to avoid introducing excessive latency - the exact heuristics for delaying in order to aggregate blocks is TBD. Tamplin & Yoshino Expires January 5, 2013 [Page 23] Internet-Draft A Multiplexing Extension for WebSockets July 2012 11. Fairness A multiplexing implementation MUST ensure reasonable fairness among the logical channels. This is accomplished in several ways: Receiver side o The receiver MAY limit the send quota of a logical channel by not replenishing it to make sure that any logical channel doesn't dominate the connection. o Send quota for one logical channel SHOULD be determined considering the processing capacity (buffer size, processing power, throughput, etc.) of that logical channel. For example, when a logical channel with excess load cannot drain data from the connection smoothly, the other logical channels get stuck even when they have room of processing capacity. Unless there's special need to give such a big quota for the channel, such condition just makes overall performance low. Sender side o The sender MUST use a fair mechanism for selecting which logical channel's data to send in the next WebSocket frame. Simple implementations may choose a round-robin scheduler, while more advanced implementations may adjust priority based on the amount or frequency of data sent by each logical channel. o The sender MUST fragment a large message into smaller frames to prevent a large message in a logical channel occupying the physical connection and thus delaying messages in other logical channels. Tamplin & Yoshino Expires January 5, 2013 [Page 24] Internet-Draft A Multiplexing Extension for WebSockets July 2012 12. Proxies Proxies which do not multiplex/demultiplex are not affected by the presence of this extension -- they simply process WebSocket frames as usual. Proxies which filter or monitor WebSocket traffic will need to understand the multiplexing extension in order to extract the data from logical connections or to terminate individual logical connections when policy is violated. Proxies which actively multiplex connections or demultiplex them (for example, a mobile network might have a proxy which aggregates WebSocket connections at a single cell to conserve bandwidth to the main gateway) will require additional configuration (perhaps including the client) that is outside the scope of this document. Tamplin & Yoshino Expires January 5, 2013 [Page 25] Internet-Draft A Multiplexing Extension for WebSockets July 2012 13. Nesting TBD: Should we allow nesting of multiplexed channels, or should we require that an intermediary multiplexing channels flatten it? The advantage of nesting is it is conceptually cleaner and less work for an intermediary, while the disadvantage is that flow control messages will get amplified by nesting and the ultimate server's job is a bit more complicated to keep a tree of channel mappings. Tamplin & Yoshino Expires January 5, 2013 [Page 26] Internet-Draft A Multiplexing Extension for WebSockets July 2012 14. Timeout When all the logical channels are closed, each endpoint MAY _Start the WebSocket Closing Handshake_ on the physical connection. Such _Start the WebSocket Closing Handshake_ operation SHOULD be delayed assuming the physical connection may be reused after some idle period. Tamplin & Yoshino Expires January 5, 2013 [Page 27] Internet-Draft A Multiplexing Extension for WebSockets July 2012 15. Close the Logical Channel To _Close the Logical Channel_, an endpoint MUST send a DropChannel multiplex control block with F bit unset. The endpoint MAY provide the reason of failure in the DropChannel block. Tamplin & Yoshino Expires January 5, 2013 [Page 28] Internet-Draft A Multiplexing Extension for WebSockets July 2012 16. Fail the Logical Channel To _Fail the Logical Channel_, an endpoint MUST send a DropChannel multiplex control block with F bit set. The endpoint MAY provide the reason of failure in the DropChannel block. Tamplin & Yoshino Expires January 5, 2013 [Page 29] Internet-Draft A Multiplexing Extension for WebSockets July 2012 17. Fail the Physical Connection To _Fail the Physical Connection_, an endpoint MUST send a DropChannel multiplex control block with objective channel ID of 0 and F bit set, and then _Fail the WebSocket Connection_ on the physical connection with status code of 1002 (TBD). Tamplin & Yoshino Expires January 5, 2013 [Page 30] Internet-Draft A Multiplexing Extension for WebSockets July 2012 18. Operations and Events on Multiplexed Connection When an endpoint is asked to perform any operation defined in the WebSocket Protocol except for _Close the WebSocket Connection_ by some application instance, the endpoint MUST perform the operation on the corresponding logical channel. If the operation involves control frames, they MUST be encapsulated in EncapsulatedControlFrame multiplex control blocks. Any event on a logical channel except for _The WebSocket Connection is Closed_, MUST be taken as one for the corresponding application instance. When an endpoint is asked to do _Close the WebSocket Connection_ by some application instance, it MUST perform _Close the Logical Channel_ on the corresponding logical channel. When a DropChannel is received, or the physical connection is closed, it MUST be taken as _The WebSocket Connection is Closed_ event for the corresponding application instance(s). What to set to _Extension In Use_ for each multiplexed connection is TBD. Tamplin & Yoshino Expires January 5, 2013 [Page 31] Internet-Draft A Multiplexing Extension for WebSockets July 2012 19. Security Considerations A client MUST be prepared to receive a NewChannelSlot with huge value on the number of slots field. Each message should consume quota for some fixed value to prohibit a channel sending lots of zero sized message to occupy the physical connection (TBD). Have upper bound for send quota. Tamplin & Yoshino Expires January 5, 2013 [Page 32] Internet-Draft A Multiplexing Extension for WebSockets July 2012 20. IANA Considerations This specification is registering a value of the Sec-WebSocket- Extension header field in accordance with Section 11.4 of the WebSocket protocol [RFC6455] as follows: Extension Identifier mux Extension Common Name Multiplexing Extension for WebSockets Extension Definition This document Known Incompatible Extensions None Tamplin & Yoshino Expires January 5, 2013 [Page 33] Internet-Draft A Multiplexing Extension for WebSockets July 2012 21. Normative References [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. [RFC6455] Fette, I. and A. Melnikov, "The WebSocket Protocol", RFC 6455, December 2011. Tamplin & Yoshino Expires January 5, 2013 [Page 34] Internet-Draft A Multiplexing Extension for WebSockets July 2012 Authors' Addresses John A. Tamplin Google, Inc. Email: jat@google.com Takeshi Yoshino Google, Inc. Email: tyoshino@google.com Tamplin & Yoshino Expires January 5, 2013 [Page 35]