Network Working Group                 Bellovin, Buchsbaum, Muthukrishnan
Internet Draft                                       AT&T Labs--Research

Expiration Date: April 2000                                 October 1999

                         TCP Compression Filter


1. Status of this Memo

   This document is an Internet-Draft and is in full conformance with
   all provisions of Section 10 of RFC2026.

   Internet-Drafts are working documents of the Internet Engineering
   Task Force (IETF), its areas, and its working groups.  Note that
   other groups may also distribute working documents as Internet-

   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 draft document will be submitted to the RFC Editor as an
   Experimental RFC. Distribution of this document is unlimited.

   The list of current Internet-Drafts can be accessed at

   The list of Internet-Draft Shadow Directories can be accessed at

2. Abstract

   We propose a TCP filter option to install compression in a virtual
   layer between TCP and the application layer.  The method is
   incrementally deployable, as neither party will install the
   compression layer without the other's consent.

Bellovin                                                FORMFEED[Page 1]

Internet Draft        draft-bellovin-tcpcomp-00.txt         October 1999

3. Introduction

   The natural place to compress data is at the application level, where
   application-specific semantics can be used to attain better
   compression.  Unfortunately, this requires changing each and every
   application, or at least changing user behavior.

   An alternative is to compress the data at the IP level, as is done in
   IPCOMP [RFC2393].  While this is application independent, its
   effectiveness is also limited, since each packet must be compressed

   We propose compressing immediately above TCP, as negotiated by a TCP
   option.  One side sends an ordered list of which compression
   algorithms it supports.  The other side selects one from the list,
   which commits both sides to compressing the payloads of all
   subsequent packets accordingly.

   An example of where this could help is the transmission of email
   messages with large attachments, often word processor documents or
   slide presentations.  Files of these types are quite compressible;
   doing the compression at a higher layer, however, would require
   either manual user intervention or changes to many different mail
   sending and receiving packages.

   This option is an example of a TCP filter option of the class
   described in [FILTDRAFT].

3.1. Specification of Requirements

   The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
   document are to be interpreted as described in RFC 2119 [RFC-2119].

Bellovin                                                FORMFEED[Page 2]

Internet Draft        draft-bellovin-tcpcomp-00.txt         October 1999

4. Option Format

   We use one TCP option, of type TCO (to be assigned by IANA), to
   signal compression.  A type field indicates the operation.

   A compression algorithm announcement MUST NOT appear except as
   specified by the [FILTDRAFT] protocol.  (All TCPs MUST ignore unknown
   options in SYN packets [RFC1122].)  Compressed packets MUST NOT be
   sent unless both parties have agreed to the appropriate filter via
   the protocol [FILTDRAFT].

   Compression algorithm IDs will be assigned by IANA.

   |  TCO   |   len  | alg...  |

   The compression filters, including any parameters, are fixed during
   the three-way handshake by the protocol [FILTDRAFT].  Subsequently,
   they may only be changed by in-band communication, i.e., by the
   compression algorithms themselves interpreting the data stream.

   |  TCO   |   len  |   alg   | parm... |

5. Behavior

   As per [FILTDRAFT], by "initiator," we indicate the party that first
   includes compression options in its SYN packet, and by "respondent,"
   we indicate the other party.

   If the respondent (cf., [FILTDRAFT] protocol) has indicated that it
   can accept a compression algorithm, a sender MUST use it.  As
   described in Section 2.2 of RFC 2393, the initiator SHOULD verify
   that compression does not increase the size of the message.  If it
   does, it SHOULD NOT initiate compression.

   Only the initial compression algorithms and parameters are determined
   by the compression options in the handshake.  Senders MAY apply
   hysteresis to sending both compressed and uncompressed packets, per
   RFC 2393, but only by using in-band communication, i.e., messages in
   the data stream itself.  In particular, to permit uncompressed data
   to be co-mingled with compressed data, we anticipate that particular
   compression algorithms will include their own header structures in
   the data stream.  Again, note that because of the stream nature of
   TCP, the uncompressed portion may be sent in the same packet as

Bellovin                                                FORMFEED[Page 3]

Internet Draft        draft-bellovin-tcpcomp-00.txt         October 1999

   compressed data.  Any necessary framing must be done by particular
   compression algorithms.  They may, however, specify the use of the
   TCP record mark filter option [DRAFTREC].

   Senders MUST honor the compression algorithm specified by the
   respondent, as per [FILTDRAFT].  Local dictates to the contrary
   require in-band communication to alter the compression behavior; if
   the compression algorithm precludes such communication, then the
   session must be terminated and re-established with different (or
   absent) compression options.

6. Interactions

6.1. TCP Urgent Pointer

   Compression filters must note application requests to send urgent
   data.  The urgent pointer passed down to TCP must point to the
   appropriate compressed bytes.  Upon receipt of an urgent packet (more
   precisely, a packet where the urgent pointer denotes a byte within
   it), the uncompression routine must send the appropriate notification
   to the application, while pointing to the proper uncompressed byte.

7. Security Considerations

   Compressing data above the TCP layer should not have any negative
   impact on security.  In particular, port numbers are not compressed.
   Some firewalls and intrusion detection systems examine TCP payload
   data, however, and they may be confused by compression.  The former
   may wish to delete the compression option; if the latter are used,
   administrators may wish to disable compression.

8. Acknowledgements

Bellovin                                                FORMFEED[Page 4]

Internet Draft        draft-bellovin-tcpcomp-00.txt         October 1999

9. References

10. Appendix

   Initial compression algorithms to be supported SHOULD include DEFLATE
   [RFC2394] (algorithm code 0x00) and LZS [RFC2395] (algorithm code

11. Author Information

Steven M. Bellovin
+1 973-360-8656

Adam L. Buchsbaum
+1 973-360-8674

S. Muthukrishnan
+1 973-360-7212

AT&T Labs--Research
Shannon Laboratory
180 Park Avenue
Florham Park, NJ 07974

Bellovin                                                FORMFEED[Page 5]