Network Working Group B. Davie
Request for Comments: 3006 C. Iturralde
Category: Standards Track D. Oran
Cisco Systems, Inc.
S. Casner
Packet Design
J. Wroclawski
MIT LCS
November 2000
Integrated Services in the Presence of Compressible Flows
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 (2000). All Rights Reserved.
Abstract
An Integrated Services (int-serv) router performs admission control
and resource allocation based on the information contained in a TSpec
(among other things). As currently defined, TSpecs convey
information about the data rate (using a token bUCket) and range of
packet sizes of the flow in question. However, the TSpec may not be
an accurate representation of the resources needed to support the
reservation if the router is able to compress the data at the link
level. This specification describes an extension to the TSpec which
enables a sender of potentially compressible data to provide hints to
int-serv routers about the compressibility they may oBTain. Routers
which support appropriate compression take advantage of the hint in
their admission control decisions and resource allocation procedures;
other routers ignore the hint. An initial application of this
approach is to notify routers performing real-time transport protocol
(RTP) header compression that they may allocate fewer resources to
RTP flows.
Table of Contents
1 Introduction ........................................... 2
2 Addition of a Hint to the Sender TSpec ................. 3
3 Admission Control and Resource Allocation .............. 4
4 Object Format .......................................... 8
4.1 Hint Numbering ......................................... 9
5 Backward Compatibility ................................. 10
6 Security Considerations ................................ 10
7 IANA Considerations .................................... 11
8 Acknowledgments ........................................ 11
9 References ............................................. 11
10 Authors' Addresses ..................................... 12
11 Full Copyright Statement ................................ 13
1. Introduction
In an Integrated Services network, RSVP [RFC2205] may be used as a
signalling protocol by which end nodes and network elements exchange
information about resource requirements, resource availability, and
the establishment and removal of resource reservations. The
Integrated Services architecture currently defines two services,
Controlled-Load [RFC2211] and Guaranteed [RFC2212]. When
establishing a reservation using either service, RSVP requires a
variety of information to be provided by the sender(s) and
receiver(s) for a particular reservation which is used for the
purposes of admission control and allocation of resources to the
reservation. Some of this information is provided by the receiver in
a FLOWSPEC object; some is provided by the sender in a SENDER_TSPEC
object [RFC2210].
A situation that is not handled well by the current specs arises when
a router that is making an admission control decision is able to
perform some sort of compression on the flow for which a reservation
is requested. For example, suppose a router is able to perform
IP/UDP/RTP header compression on one of its interfaces [RFC2508].
The bandwidth needed to accommodate a compressible flow on that
interface would be less than the amount contained in the
SENDER_TSPEC. Thus the router might erroneously reject a reservation
that could in fact have been accommodated. At the same time, the
sender is not at liberty to reduce its TSpec to account for the
compression of the data, since it does not know if the routers along
the path are in fact able to perform compression. Furthermore, it is
probable that only a subset of the routers on the path (e.g., those
connected to low-speed serial links) will perform compression.
This specification describes a mechanism by which the sender can
provide a hint to network elements regarding the compressibility of
the data stream that it will generate. Network elements may use this
hint as an additional piece of data when making admission control and
resource allocation decisions.
This specification is restricted to the case where compression is
performed only on a link-by-link basis, as with header compression.
Other cases (e.g., transcoding, audio silence detection) which would
affect the bandwidth consumed at all downstream nodes are for further
study. In these latter cases, it would be necessary to modify a
sender TSpec as it is passed through a compressing node. In the
approach presented here, the sender TSpec that appears on the wire is
never modified, just as specified in [RFC2210].
2. Addition of a Hint to the Sender TSpec
The appropriate place for a `compressibility hint' is the Sender
TSpec. The reasons for this choice are:
- The sender is the party who knows best what the data will look
like.
- Unlike the Adspec, the Sender TSpec is not modified in transit
- From the perspective of RSVP, the Sender TSpec is a set of
opaque parameters that are passed to `traffic control'
(admission control and resource allocation); the
compressibility hint is just such a parameter.
An alternative to putting this information in the TSpec would be to
use an additional object in the RSVP PATH message. While this could
be made to work for RSVP, it does not address the issue of how to get
the same information to an intserv router when mechanisms other than
RSVP are used to reserve resources. It would also imply a change to
RSVP message processing just for the purposes of getting more
information to entities that are logically not part of RSVP
(admission control and resource allocation). The inclusion of the
information in the TSpec seems preferable and more consistent with
the Integrated Services architecture.
The contents of the hint are likely to vary depending on the exact
scenario. The hint needs to tell the routers that receive it:
- the type of compression that is possible on this flow (e.g.
IP/UDP/RTP);
- enough information to enable a router to determine the likely
compression ratio that may be achieved.
In a simple case such as IP/UDP/RTP header compression, it may be
sufficient to tell the routers nothing more than the fact that
IP/UDP/RTP data is being sent. Knowing this fact, the maximum packet
size of the flow (from the TSpec), and the local conditions at the
router, may be sufficient to allow the router to determine the
reduction in bandwidth that compression will allow. In other cases,
it may be helpful or necessary for the sender to include additional
quantitative information to assist in the calculation of the
compression ratio. To handle these cases, additional parameters
containing various amounts of information may be added to the sender
TSpec. Details of the encoding of these parameters, following the
approach originally described in [RFC2210] are described below.
3. Admission Control and Resource Allocation
Integrated Services routers make admission control and resource
allocation decisions based on, among other things, information in the
sender TSpec. If a router receives a sender TSpec which contains a
compressibility hint, it may use the hint to calculate a `compressed
TSpec' which can be used as input to the admission control and
resource allocation processes in place of the TSpec provided by the
sender. To make this concrete, consider the following simple
example. A router receives a reservation request for controlled load
service where:
- The Sender TSpec and Receiver TSpec contain identical token
bucket parameters;
- The rate parameter in the token bucket (r) is 48 kbps;
- The token bucket depth (b) is 120 bytes;
- The maximum packet size (M) in the TSpecs is 120 bytes;
- The minimum policed unit (m) is 64 bytes;
- The Sender TSpec contains a compressibility hint indicating
that the data is IP/UDP/RTP;
- The compressibility hint includes a compression factor of 70%,
meaning that IP/UDP/RTP header compression will cause a
reduction in bandwidth consumed at the link level by a factor
of 0.7 (the result of compressing 40 bytes of IP/UDP/RTP header
to 4 bytes on a 120 byte packet)
- The interface on which the reservation is to be installed is
able to perform IP/UDP/RTP header compression.
The router may thus conclude that it can scale down the token bucket
parameters r and b by a factor of 0.7, i.e., to 33.6 kbps and 84
bytes respectively. M may be scaled down by the same factor (to 84
bytes), but a different calculation should be used for m. If the
sender actually sends a packet of size m, its header may be
compressed from 40 bytes to 4, thus reducing the packet to 28 bytes;
this value should be used for m.
Note that if the source always sends packets of the same size and
IP/UDP/RTP always works perfectly, the compression factor is not
strictly needed. The router can independently determine that it can
compress the 40 bytes of IP/UDP/RTP header to 4 bytes (with high
probability). To determine the worst-case (smallest) gain provided
by compression, it can assume that the sender always sends maximum
sized packets at 48 kbps, i.e., a 120 byte packet every 20
milliseconds. The router can conclude that these packets would be
compressed to 84 bytes, yielding a token bucket rate of 33.6 kbps and
a token bucket depth of 84 bytes as before. If the sender is willing
to allow an independent calculation of compression gain by the
router, the eXPlicit compression factor may be omitted from the
TSpec. Details of the TSpec encoding are provided below.
To generalize the above discussion, assume that the Sender TSpec
consists of values (r, b, p, M, m), that the explicit compression
factor provided by the sender is f percent, and that the number of
bytes saved by compression is N, independent of packet size. The
parameters in the compressed TSpec would be:
r' = r * f/100
b' = b * f/100
p' = p
M' = M-N
m' = m-N
The calculations for r' and b' reflect that fact that f is expressed
as a percentage and must therefore be divided by 100. The
calculations for M' and m' hold only in the case where the
compression algorithm reduces packets by a certain number of bytes
independent of content or length of the packet, as is true for header
compression. Other compression algorithms may not have this
property. In determining the value of N, the router may need to make
worst case assumptions about the number of bytes that may be removed
by compression, which depends on such factors as the presence of UDP
checksums and the linearity of RTP timestamps.
All these adjusted values are used in the compressed TSpec. The
router's admission control and resource allocation algorithms should
behave as if the sender TSpec contained those values. [RFC2205]
provides a set of rules by which sender and receiver TSpecs are
combined to calculate a single `effective' TSpec that is passed to
admission control. When a reservation covering multiple senders is
to be installed, it is necessary to reduce each sender TSpec by its
appropriate compression factor. The set of sender TSpecs that apply
to a single reservation on an interface are added together to form
the effective sender TSpec, which is passed to traffic control. The
effective receiver TSpec need not be modified; traffic control takes
the greatest lower bound of these two TSpecs when making its
admission control and resource allocation decisions.
The handling of the receiver RSpec depends on whether controlled load
or guaranteed service is used. In the case of controlled load, no
additional processing of RSpec is needed. However, a guaranteed
service RSpec contains a rate term R which does need to be adjusted
downwards to account for compression. To determine how R should be
adjusted, we note that the receiver has chosen R to meet a certain
delay goal, and that the terms in the delay equation that depend on R
are b/R and C/R (when the peak rate is large). The burstsize b in
this case is the sum of the burstsizes of all the senders for this
reservation, and each of these numbers has been scaled down by the
appropriate compression factor. Thus, R should be scaled down using
an average compression factor
f_avg = (b1*f1 + b2*f2 + ... + bn*fn)/(b1 + b2 + ... bn)
where bk is the burstsize of sender k and fk is the corresponding
compression factor for this sender. Note that f_avg, like the
individual fi's, is a percentage. Note also that this results in a
compression factor of f in the case where all senders use the same
compression factor f.
To prevent an increase in delay caused by the C/R term when the
reduced value of R is used for the reservation, it is necessary for
this hop to `inflate' its value of C by dividing it by (f_avg/100).
This will cause the contribution to delay made by this hop's C term
to be what the receiver would expect when it chooses its value of R.
There are certain risks in adjusting the resource requirements
downwards for the purposes of admission control and resource
allocation. Most compression algorithms are not completely
deterministic, and thus there is a risk that a flow will turn out to
be less compressible than had been assumed by admission control.
This risk is reduced by the use of the explicit compression factor
provided by the sender, and may be minimized if the router makes
worst case assumptions about the amount of compression that may be
achieved. This is somewhat analogous to the tradeoff between making
worst case assumptions when performing admission control or making
more optimistic assumptions, as in the case of measurement-based
admission control. If a flow turns out to be less compressible that
had been assumed when performing admission control, any extra traffic
will need to be policed according to normal intserv rules. For
example, if the router assumed that the 48 kbps stream above could be
compressed to 33.6 kbps and it was ultimately possible to compress it
to 35 kbps, the extra 1.4 kbps would be treated as excess. The exact
treatment of such excess is service dependent.
A similar scenario may arise if a sender claims that data for a
certain session is compressible when in fact it is not, or overstates
the extent of its compressibility. This might cause the flow to be
erroneously admitted, and would cause insufficient resources to be
allocated to it. To prevent such behavior from adversely affecting
other reserved flows, any flow that sends a compressibility hint
should be policed (in any router that has made use of the hint for
its admission control) on the assumption that it is indeed
compressible, i.e., using the compressed TSpec. That is, if the flow
is found to be less compressible than advertised, the extra traffic
that must be forwarded by the router above the compressed TSpec will
be policed according to intserv rules appropriate for the service.
Note that services that use the maximum datagram size M for policing
purposes (e.g. guaranteed service [RFC2210]) should continue to use
the uncompressed value of M to allow for the possibility that some
packets may not be successfully compressed.
Note that RSVP does not generally require flows to be policed at
every hop. To quote [RFC2205]:
Some QoS services may require traffic policing at some or all of
(1) the edge of the network, (2) a merging point for data from
multiple senders, and/or (3) a branch point where traffic flow
from upstream may be greater than the downstream reservation being
requested. RSVP knows where such points occur and must so
indicate to the traffic control mechanism.
For the purposes of policing, a router which makes use of the
compressibility hint in a sender TSpec should behave as if it is at
the edge of the network, because it is in a position to receive
traffic from a sender that, while it passed through policing at the
real network edge, may still need to be policed if the amount of data
sent exceeds the amount described by the compressed TSpec.
4. Object Format
The compressibility hint may be included in the sender TSpec using
the encoding rules of Appendix A in [RFC2210]. The complete sender
TSpec is as follows:
31 24 23 16 15 8 7 0
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
1 0 (a) reserved 10 (b)
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
2 1 (c) 0 reserved 9 (d)
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
3 127 (e) 0 (f) 5 (g)
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
4 Token Bucket Rate [r] (32-bit IEEE floating point number)
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
5 Token Bucket Size [b] (32-bit IEEE floating point number)
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
6 Peak Data Rate [p] (32-bit IEEE floating point number)
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
7 Minimum Policed Unit [m] (32-bit integer)
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
8 Maximum Packet Size [M] (32-bit integer)
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
9 126 (h) 0 (i) 2 (j)
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
10 Hint (assigned number)
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
11 Compression factor [f] (32-bit integer)
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
(a) - Message format version number (0)
(b) - Overall length (10 Words not including header)
(c) - Service header, service number 1 (default/global
information)
(d) - Length of service 1 data, 9 words not including header
(e) - Parameter ID, parameter 127 (Token_Bucket_TSpec)
(f) - Parameter 127 flags (none set)
(g) - Parameter 127 length, 5 words not including header
(h) - Parameter ID, parameter 126 (Compression_Hint)
(i) - Parameter 126 flags (none set)
(j) - Parameter 126 length, 2 words not including header
The difference between this TSpec and the one described in [RFC2210]
is that the overall length contained in the first word is increased
by 3, as is the length of the `service 1 data', and the original
TSpec parameters are followed by a new parameter, the compressibility
hint. This parameter contains the standard parameter header, and an
assigned number indicating the type of compression that is possible
on this data. Different values of the hint would imply different
compression algorithms may be applied to the data. Details of the
numbering scheme for hints appear below.
Following the hint value is the compression factor f, expressed as a
32 bit integer representing the factor as a percentage value. The
valid range for this factor is (0,100]. A sender that does not know
what value to use here or wishes to leave the compression factor
calculation to the routers' discretion may use the reserved value 0
to indicate this fact. Zero is reserved because it is not possible
to compress a data stream to zero bits per second. The value 100
indicates that no compression is expected on this stream.
In some cases, additional quantitative information about the traffic
may be required to enable a router to determine the amount of
compression possible. In this case, a different encoding of the
parameter would be required.
In some cases it may be desirable to include more than one hint in a
Tspec (e.g., because more than one compression scheme could be
applied to the data.) In this case, multiple instances of parameter
126 may appear in the Tspec and the overall length of the Tspec and
the length of the Service 1 data would be increased accordingly.
Note that the Compression_Hint is, like the Token_Bucket_Tspec, not
specific to a single service, and thus has a parameter value less
than 128. It is also included as part of the default/global
information (service number 1).
4.1. Hint Numbering
Hints are represented by a 32 bit field, with the high order 16 bits
being the IP-compression-protocol number as defined in [RFC1332] and
[RFC2509]. The low order 16 bits are a sub-option for the cases
where the IP-compression-protocol number alone is not sufficient for
int-serv purposes. The following hint values are required at the
time of writing:
- hint = 0x002d0000: IP/TCP data that may be compressed according
to [RFC1144]
- hint = 0x00610000: IP data that may be compressed according to
[RFC2507]
- hint = 0x00610100: IP/UDP/RTP data that may be compressed
according to [RFC2508]
5. Backward Compatibility
It is desirable that an intserv router which receives this new TSpec
format and does not understand the compressibility hint should
silently ignore the hint rather than rejecting the entire TSpec (or
the message containing it) as malformed. While [RFC2210] clearly
specifies the format of TSpecs in a way that they can be parsed even
when they contain unknown parameters, it does not specify what action
should be taken when unknown objects are received. Thus it is quite
possible that some RSVP implementations will discard PATH messages
containing a TSpec with the compressibility hint. In such a case,
the router should send a PathErr message to the sending host. The
message should indicate a malformed TSpec (Error code 21, Sub-code
04). The host may conclude that the hint caused the problem and send
a new PATH without the hint.
For the purposes of this specification, it would be preferable if
unknown TSpec parameters could be silently ignored. In the case
where a parameter is silently ignored, the node should behave as if
that parameter were not present, but leave the unknown parameter
intact in the object that it forwards. This should be the default
for unknown parameters of the type described in [RFC2210].
It is possible that some future modifications to [RFC2210] will
require unknown parameter types to cause an error response. This
situation is analogous to RSVP's handling of unknown objects, which
allows for three different response to an unknown object, based on
the highest two bits of the Class-Num. One way to handle this would
be to divide the parameter space further than already done in [RFC
2216]. For example, parameter numbers of the form x1xxxxxx could be
silently ignored if unrecognized, while parameter numbers of the form
x0xxxxxx could cause an error response if unrecognized. (The meaning
of the highest order bit is already fixed by [RFC2216].) A third
possibility exists, which is to remove the unrecognized parameter
before forwarding, but this does not seem to be useful.
6. Security Considerations
The extensions defined in this document pose essentially the same
security risks as those of [RFC2210]. The risk that a sender will
falsely declare his data to be compressible is equivalent to the
sender providing an insufficiently large TSpec and is dealt with in
the same way.
7. IANA Considerations
This specification relies on IANA-assigned numbers for the
compression scheme hint. Where possible the existing numbering
scheme for compression algorithm identification in PPP has been used,
but it may in the future be necessary for IANA to assign hint numbers
purely for the purposes of int-serv.
8. Acknowledgments
Carsten Borman and Mike DiBiasio provided much helpful feedback on
this document.
9. References
[RFC1144] Jacobson, V., "Compressing TCP/IP Headers for Low-Speed
Serial Links", RFC1144, February 1990.
[RFC1332] McGregor, G., "The PPP Internet Protocol Control Protocol
(IPCP)", RFC1332, May 1992.
[RFC2205] Braden, R., Zhang, L., Berson, S., Herzog, S. and S.
Jamin, "Resource ReSerVation Protocol (RSVP) -- Version 1
Functional Specification", RFC2205, September 1997.
[RFC2210] Wroclawski, J., "The Use of RSVP with IETF Integrated
Services", RFC2210, September 1997.
[RFC2211] Wroclawski, J., "Specification of the Controlled-Load
Network Element Service", RFC2211, September 1997.
[RFC2212] Shenker, S., Partridge, C. and R. Guerin, "Specification
of Guaranteed Quality of Service", RFC2212, September
1997.
[RFC2216] Shenker, S. and J. Wroclawski, "Network Element Service
Specification Template", RFC2216, September 1997.
[RFC2507] Degermark, M., Nordgren, B. and S. Pink,"Header
Compression for IP", RFC2507, February 1999.
[RFC2508] Casner, S. and V. Jacobson, "Compressing IP/UDP/RTP
Headers for Low-Speed Serial Links", RFC2508, February
1999.
[RFC2509] Engan, M., Casner, S. and C. Bormann, "IP Header
Compression over PPP", RFC2509, February 1999.
10. Authors' Addresses
Bruce Davie
Cisco Systems, Inc.
250 Apollo Drive
Chelmsford, MA, 01824
EMail: bsd@cisco.com
Carol Iturralde
Cisco Systems, Inc.
250 Apollo Drive
Chelmsford, MA, 01824
EMail: cei@cisco.com
Dave Oran
Cisco Systems, Inc.
170 Tasman Drive
San Jose, CA, 95134
EMail: oran@cisco.com
Stephen L. Casner
Packet Design
66 Willow Place
Menlo Park, CA 94025
EMail: casner@acm.org
John Wroclawski
MIT Laboratory for Computer Science
545 Technology Sq.
Cambridge, MA 02139
EMail: jtw@lcs.mit.edu
Full Copyright Statement
Copyright (C) The Internet Society (2000). 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 RFCEditor function is currently provided by the
Internet Society.
