BEEP over TCP: Mapping the BEEP Core onto TCP
How a BEEP session runs on one TCP connection, and the window-based flow control BEEP adds on top so channels do not starve each other.
Rose, M., "Mapping the BEEP Core onto TCP", RFC 3081, DOI 10.17487/RFC3081, March 2001, <https://www.rfc-editor.org/info/rfc3081>.
@misc{rfc3081,
series = {Request for Comments},
number = 3081,
howpublished = {RFC 3081},
publisher = {RFC Editor},
doi = {10.17487/RFC3081},
url = {https://www.rfc-editor.org/info/rfc3081},
author = {M. Rose},
title = {{Mapping the BEEP Core onto TCP}},
pagetotal = 8,
year = 2001,
month = mar,
} RFC text
Contents
Network Working Group M. Rose
Request for Comments: 3081 Invisible Worlds, Inc.
Category: Standards Track March 2001
Mapping the BEEP Core onto TCP
Status of this Memo
This document specifies an Internet standards track protocol for the
Internet community, and requests discussion and suggestions for
improvements. Please refer to the current edition of the "Internet
Official Protocol Standards" (STD 1) for the standardization state
and status of this protocol. Distribution of this memo is unlimited.
Copyright Notice
Copyright (C) The Internet Society (2001). All Rights Reserved.
Abstract
This memo describes how a BEEP (Blocks Extensible Exchange Protocol)
session is mapped onto a single TCP (Transmission Control Protocol)
connection.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 1
2. Session Management . . . . . . . . . . . . . . . . . . . . . 2
3. Message Exchange . . . . . . . . . . . . . . . . . . . . . . 2
3.1 Flow Control . . . . . . . . . . . . . . . . . . . . . . . . 3
3.1.1 Channel Creation . . . . . . . . . . . . . . . . . . . . . . 3
3.1.2 Sending Messages . . . . . . . . . . . . . . . . . . . . . . 3
3.1.3 Processing SEQ Frames . . . . . . . . . . . . . . . . . . . 4
3.1.4 Use of Flow Control . . . . . . . . . . . . . . . . . . . . 4
4. Security Considerations . . . . . . . . . . . . . . . . . . 6
References . . . . . . . . . . . . . . . . . . . . . . . . . 6
Author's Address . . . . . . . . . . . . . . . . . . . . . . 6
A. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 7
Full Copyright Statement . . . . . . . . . . . . . . . . . . 8
1. Introduction
This memo describes how a BEEP [1] session is mapped onto a single
TCP [2] connection. Refer to Section 2.5 of [1] for an explanation
of the mapping requirements.
2. Session Management
The mapping of BEEP session management onto the TCP service is
straight-forward.
A BEEP session is established when a TCP connection is established
between two BEEP peers:
o the BEEP peer that issues a passive TCP OPEN call is termed the
listener; and,
o the BEEP peer that issues an active TCP OPEN call is termed the
initiator.
A simultaneous TCP OPEN would result in both BEEP peers believing
they are the initiator and neither peer will be able to start any
channels. Because of this, services based on BEEP must be designed
so that simultaneous TCP OPENs cannot occur.
If both peers agree to release a BEEP session (c.f., [1]'s Section
2.4), the peer sending the "ok" reply, immediately issues the TCP
CLOSE call. Upon receiving the reply, the other peer immediately
issues the TCP CLOSE call.
A BEEP session is terminated when either peer issues the TCP ABORT
call, and the TCP connection is subsequently aborted.
3. Message Exchange
The mapping of BEEP exchanges onto the TCP service is less straight-
forward.
Messages are reliably sent and received using TCP's SEND and RECEIVE
calls. (This also provides ordered delivery of messages on the same
channel.)
Although TCP imposes flow control on a per-connection basis, if
multiple channels are simultaneously in use on a BEEP session, BEEP
must provide a mechanism to avoid starvation and deadlock. To
achieve this, BEEP re-introduces a mechanism used by the TCP:
window-based flow control -- each channel has a sliding window that
indicates the number of payload octets that a peer may transmit
before receiving further permission.
3.1 Flow Control
Recall from Section 2.2.1.2 of [1] that every payload octet sent in
each direction on a channel has an associated sequence number.
Numbering of payload octets within a data frame is such that the
first payload octet is the lowest numbered, and the following payload
octets are numbered consecutively.
The actual sequence number space is finite, though very large,
ranging from 0..4294967295 (2**32 - 1). Since the space is finite,
all arithmetic dealing with sequence numbers is performed modulo
2**32. This unsigned arithmetic preserves the relationship of
sequence numbers as they cycle from 2**32 - 1 to 0 again. Consult
Sections 2 through 5 of [3] for a discussion of the arithmetic
properties of sequence numbers.
3.1.1 Channel Creation
When a channel is created, the sequence number associated with the
first payload octet of the first data frame is 0, and the initial
window size for that channel is 4096 octets. After channel creation,
a BEEP peer may update the window size by sending a SEQ frame
(Section 3.1.3).
If a BEEP peer is asked to create a channel and it is unable to
allocate at least 4096 octets for that channel, it must decline
creation of the channel, as specified in Section 2.3.1.2 of [1].
Similarly, during establishment of the BEEP session, if the BEEP peer
acting in the listening role is unable to allocate at least 4096
octets for channel 0, then it must return a negative reply, as
specified in Section 2.4 of [1], instead of a greeting.
3.1.2 Sending Messages
Before a message is sent, the sending BEEP peer must ensure that the
size of the payload is within the window advertised by the receiving
BEEP peer. If not, it has three choices:
o if the window would allow for at least one payload octet to be
sent, the BEEP peer may segment the message and start by sending a
smaller data frame (up to the size of the remaining window);
o the BEEP peer may delay sending the message until the window
becomes larger; or,
o the BEEP peer may signal to its application that it is unable to
send the message, allowing the application to try again at a later
time (or perhaps signaling its application when a larger window is
available).
The choice is implementation-dependent, although it is recommended
that the application using BEEP be given a mechanism for influencing
the decision.
3.1.3 Processing SEQ Frames
As an application accepts responsibility for incoming data frames,
its BEEP peer should send SEQ frames to advertise a new window.
The ABNF [4] for a SEQ frame is:
seq = "SEQ" SP channel SP ackno SP window CR LF
ackno = seqno
window = size
; channel, seqno, and size are defined in Section 2.2.1 of [1].
The SEQ frame has three parameters:
o a channel number;
o an acknowledgement number, that indicates the value of the next
sequence number that the sender is expecting to receive on this
channel; and,
o a window size, that indicates the number of payload octets
beginning with the one indicated by the acknowledgement number
that the sender is expecting to receive on this channel.
A single space character (decimal code 32, " ") separates each
component. The SEQ frame is terminated with a CRLF pair.
When a SEQ frame is received, if any of the channel number,
acknowledgement number, or window size cannot be determined or is
invalid, then the BEEP session is terminated without generating a
response, and it is recommended that a diagnostic entry be logged.
3.1.4 Use of Flow Control
The key to successful use of flow control within BEEP is to balance
performance and fairness:
o large messages should be segmented into frames no larger than
two-thirds of TCP's negotiated maximum segment size;
o frames for different channels with traffic ready to send should be
sent in a round-robin fashion;
o each time a frame is received, a SEQ frame should be sent whenever
the window size that will be sent is at least one half of the
buffer space available to this channel; and,
o if the transport service presents multiple frames to a BEEP peer
simultaneously, then a single consolidating SEQ frame may be sent.
In order to avoid pathological interactions with the transport
service, it is important that a BEEP peer advertise windows based on
available buffer space, to allow data to be read from the transport
service as soon as available. Further, SEQ frames for a channel must
have higher priority than messages for that channel.
Implementations may wish to provide queue management facilities to
the application using BEEP, e.g., channel priorities, (relative)
buffer allocations, and so on. In particular, implementations should
not allow a given channel to monopolize the underlying transport
window (e.g., slow readers should get small windows).
In addition, where possible, implementations should support transport
layer APIs that convey congestion information. These APIs allow an
implementation to determine its share of the available bandwidth, and
also be notified of changes in the estimated path bandwidth. Note
that when a BEEP session has multiple channels that are
simultaneously exchanging large messages, implementations without
access to this information may have uncertain fairness and progress
properties during times of network congestion.
Finally, implementors should follow the guidelines given in the
relevant portions of RFC1122 [5] that deal with flow control (and
bear in mind that issues such as retransmission, while they interact
with flow control in TCP, are not applicable to this memo). For
example, Section 4.2.2.16 of RFC1122 [5] indicates that a "receiver
SHOULD NOT shrink the window, i.e., move the right window edge to the
left" and then discusses the impact of this rule on unacknowledged
data. In the context of mapping BEEP onto a single TCP connection,
only the portions concerning flow control should be implemented.
4. Security Considerations
Consult Section [1]'s Section 9 for a discussion of security issues.
References
[1] Rose, M., "The Blocks Extensible Exchange Protocol Core", RFC
3080, March 2001.
[2] Postel, J., "Transmission Control Protocol", STD 7, RFC 793,
September 1981.
[3] Elz, R. and R. Bush, "Serial Number Arithmetic", RFC 1982,
August 1996.
[4] Crocker, D. and P. Overell, "Augmented BNF for Syntax
Specifications: ABNF", RFC 2234, November 1997.
[5] Braden, R., "Requirements for Internet Hosts -- Communication
Layers", STD 3, RFC 1122, October 1989.
Author's Address
Marshall T. Rose
Invisible Worlds, Inc.
1179 North McDowell Boulevard
Petaluma, CA 94954-6559
US
Phone: +1 707 789 3700
EMail: mrose@invisible.net
URI: http://invisible.net/
Appendix A. Acknowledgements
The author gratefully acknowledges the contributions of: Dave
Crocker, Steve Harris, Eliot Lear, Keith McCloghrie, Craig Partridge,
Vernon Schryver, and, Joe Touch. In particular, Dave Crocker
provided helpful suggestions on the nature of flow control in the
mapping.
Full Copyright Statement
Copyright (C) The Internet Society (2001). All Rights Reserved.
This document and translations of it may be copied and furnished to
others, and derivative works that comment on or otherwise explain it
or assist in its implementation may be prepared, copied, published
and distributed, in whole or in part, without restriction of any
kind, provided that the above copyright notice and this paragraph are
included on all such copies and derivative works. However, this
document itself may not be modified in any way, such as by removing
the copyright notice or references to the Internet Society or other
Internet organizations, except as needed for the purpose of
developing Internet standards in which case the procedures for
copyrights defined in the Internet Standards process must be
followed, or as required to translate it into languages other than
English.
The limited permissions granted above are perpetual and will not be
revoked by the Internet Society or its successors or assigns.
This document and the information contained herein is provided on an
"AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING
TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING
BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION
HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF
MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
Acknowledgement
Funding for the RFC Editor function is currently provided by the
Internet Society.
AI ELI5
On this page
§1What This Mapping Defines
BEEP (Blocks Extensible Exchange Protocol) is defined in RFC 3080. It needs an underlying transport, and this memo says how a BEEP session is mapped onto a single TCP connection. The mapping requirements themselves are explained in Section 2.5 of the BEEP core specification.
The RFC is short. It covers session management (opening and closing) and message exchange, where most of the work is flow control.
BEEP is a way for programs to chat. This note explains how to carry the chat on the usual reliable Internet connection, TCP.
§2Session Management
The mapping is straightforward. A BEEP session begins when a TCP connection is established between two BEEP peers. The peer that issues a passive TCP OPEN is the listener, and the one that issues an active TCP OPEN is the initiator.
A simultaneous TCP OPEN would make both peers believe they are the initiator, and neither could start any channels. Services based on BEEP must therefore be designed so that simultaneous opens cannot occur.
To release a session, the peer sending the "ok" reply immediately issues TCP CLOSE, and the other peer does the same on receiving the reply. A session is terminated when either peer issues TCP ABORT.
Whoever picks up the phone is the listener, and whoever dials is the initiator. If both dial at once, nobody knows who is who, so the design must avoid that.
§3Message Exchange and Why Flow Control Is Needed
Messages are sent and received reliably with TCP's SEND and RECEIVE calls, which also gives ordered delivery on the same channel. TCP flow control works per connection, though. When several channels are in use at once, BEEP must prevent starvation and deadlock itself.
So BEEP reintroduces a TCP idea: window-based flow control. Each channel has a sliding window giving the number of payload octets a peer may send before getting further permission.
Every payload octet in each direction on a channel has a sequence number, with the first octet of a data frame the lowest. The space runs from 0 to 4294967295, and arithmetic is done modulo 232.
If one kid talks nonstop on a shared phone line, the others never get a turn. So each kid gets a limit on how much they can say before being allowed to say more.
§3.1Windows and Sending Messages
When a channel is created, the first payload octet has sequence number 0 and the initial window is 4096 octets. If a peer cannot allocate at least 4096 octets for a new channel, it must decline the channel. If the listener cannot allocate 4096 octets for channel 0, it must return a negative reply instead of a greeting.
Before sending a message, the sender must make sure the payload fits in the window the receiver advertised. If it does not, it has three choices:
- Segment the message and send a smaller frame, if the window allows at least one octet.
- Delay the message until the window grows.
- Tell its application it cannot send now, so the application can retry later.
The choice is implementation-dependent, but the application should be given a way to influence it.
Each channel starts with room for 4096 octets of talking. If your message is too big, you send part of it, wait, or say "try later".
§3.1.3SEQ Frames
As an application takes responsibility for incoming data, its BEEP peer should send a SEQ frame to advertise a new window. In ABNF, a frame is "SEQ" SP channel SP ackno SP window CR LF.
The three parameters are the channel number, an acknowledgement number (the next sequence number the sender expects to receive on that channel) and a window size (how many payload octets, starting at the acknowledgement number, the sender expects to receive). A single space separates the parts, and the frame ends with a CRLF pair.
If the channel number, acknowledgement number or window size cannot be determined or is invalid, the BEEP session is terminated without a response, and a diagnostic entry should be logged.
A SEQ frame is a short note saying "I have read up to here, and you may send me this much more."
seq = "SEQ" SP channel SP ackno SP window CR LF
ackno = seqno
window = size§3.1.4Balancing Performance and Fairness
The RFC gives guidance for balancing speed and fairness:
- Segment large messages into frames no larger than two-thirds of TCP's negotiated maximum segment size.
- Send frames for channels with traffic ready in a round-robin fashion.
- Send a
SEQframe after receiving a frame whenever the window to be sent is at least one half of the buffer space available to the channel. - If several frames arrive at once, a single consolidating
SEQframe may be sent.
Windows should be based on available buffer space, and SEQ frames for a channel must have higher priority than messages for that channel. No channel should monopolize the transport window, so slow readers should get small windows.
Implementations should support transport APIs that convey congestion information where possible. Without them, fairness and progress during congestion may be uncertain. Implementers should also follow the flow control parts of RFC 1122, for example that a receiver SHOULD NOT shrink the window, but only the flow control portions, since retransmission does not apply here.
Take turns, send small pieces, and only hand out more room when you really have space for it. A slow reader should get a small allowance.
§4Security Considerations
This RFC adds no security text of its own. It points the reader to Section 9 of the BEEP core specification, RFC 3080, for the discussion of security issues.
The only security-related behavior defined here is that a SEQ frame with an undeterminable or invalid parameter ends the session.
The safety rules live in the main BEEP document, so this one just points there.