Network Working Group K. Holtman
Request for Comments: 2296 TUE
Category: EXPerimental A. Mutz
Hewlett-Packard
March 1998

HTTP Remote Variant Selection Algorithm -- RVSA/1.0

Status of this Memo

This memo defines an Experimental Protocol for the Internet
community. It does not specify an Internet standard of any kind.
Discussion and suggestions for improvement are requested.
Distribution of this memo is unlimited.

Copyright Notice

Copyright (C) The Internet Society (1998). All Rights Reserved.

ABSTRACT

HTTP allows web site authors to put multiple versions of the same
information under a single URL. Transparent content negotiation is a
mechanism for automatically selecting the best version when the URL
is Accessed. A remote variant selection algorithm can be used to
speed up the transparent negotiation process. This document defines
the remote variant selection algorithm with the version number 1.0.

TABLE OF CONTENTS

1 IntrodUCtion...............................................2
2 Terminology and notation...................................2
3 The remote variant selection algorithm.....................2
3.1 Input....................................................2
3.2 Output...................................................3
3.3 Computing overall quality values.........................3
3.4 Definite and speculative quality values..................5
3.5 Determining the result...................................6
4 Use of the algorithm.......................................7
4.1 Using quality factors to rank preferences................7
4.2 Construction of short requests...........................8
4.2.1 Collapsing Accept- header elements.....................8
4.2.2 Omitting Accept- headers...............................9
4.2.3 Dynamically lengthening requests.......................9
4.3 Differences between the local and the remote algorithm..10
4.3.1 Avoiding major differences............................11
4.3.2 Working around minor differences......................11

5 Security and privacy considerations.......................11
6 Acknowledgments...........................................12
7 References................................................12
8 Authors' Addresses........................................12
9 Full Copyright Statement..................................13

1 Introduction

HTTP allows web site authors to put multiple versions (variants) of
the same information under a single URL. Transparent content
negotiation [2] is a mechanism for automatically selecting the best
variant when the URL is accessed. A remote variant selection
algorithm can be used by a HTTP server to choose a best variant on
behalf of a negotiating user agent. The use of a remote algorithm
can speed up the transparent negotiation process by eliminating a
request-response round trip.

This document defines the remote variant selection algorithm with the
version number 1.0. The algorithm computes whether the Accept-
headers in the request contain sufficient information to allow a
choice, and if so, which variant must be chosen.

2 Terminology and notation

This specification uses the terminology and notation of the HTTP
transparent content negotiation specification [2].

3 The remote variant selection algorithm

This section defines the remote variant selection algorithm with the
version number 1.0. To implement this definition, a server MAY run
any algorithm which gives equal results.

Note: According to [2], servers are always free to return a list
response instead of running a remote algorithm. Therefore,
whenever a server may run a remote algorithm, it may also run a
partial implementation of the algorithm, provided that the partial
implementation always returns List_response when it cannot compute
the real result.

3.1 Input

The algorithm is always run for a particular request on a
particular transparently negotiable resource. It takes the
following information as input.

1. The variant list of the resource, as present in the Alternates
header of the resource.

2. (Partial) Information about capabilities and preferences of the
user agent for this particular request, as given in the Accept-
headers of the request.

If a fallback variant description

{"fallback.Html"}

is present in the Alternates header, the algorithm MUST interpret it
as the variant description

{"fallback.html" 0.000001}

The extremely low source quality value ensures that the fallback
variant only gets chosen if all other options are exhausted.

3.2 Output

As its output, the remote variant selection algorithm and will yield
the appropriate action to be performed. There are two possibilities:

Choice_response

The Accept- headers contain sufficient information to make a
choice on behalf of the user agent possible, and the best
variant MAY be returned in a choice response.

List_response

The Accept- headers do not contain sufficient information to
make a choice on behalf of the user agent possible. A list
response MUST be returned, allowing the user agent to make the
choice itself.

3.3 Computing overall quality values

As a first step in the remote variant selection algorithm, the
overall qualities of the individual variants in the list are
computed.

The overall quality Q of a variant is the value

Q = round5( qs * qt * qc * ql * qf )

where round5 is a function which rounds a floating point value to 5
decimal places after the point, and where the factors qs, qt, qc, ql,
and qf are determined as follows.

qs Is the source quality factor in the variant description.

qt The media type quality factor is 1 if there is no type
attribute in the variant description, or if there is no Accept
header in the request. Otherwise, it is the quality assigned
by the Accept header to the media type in the type attribute.

Note: If a type is matched by none of the elements of an
Accept header, the Accept header assigns the quality factor 0
to that type.

qc The charset quality factor is 1 if there is no charset
attribute in the variant description, or if there is no
Accept-Charset header in the request. Otherwise, the charset
quality factor is the quality assigned by the Accept-Charset
header to the charset in the charset attribute.

ql The language quality factor is 1 if there is no language
attribute in the variant description, or if there is no
Accept-Language header in the request. Otherwise, the language
quality factor is the highest quality factor assigned by the
Accept-Language header to any one of the languages listed in
the language attribute.

qf The features quality factor is 1 if there is no features
attribute in the variant description, or if there is no
Accept-Features header in the request. Otherwise, it is the
quality degradation factor for the features attribute, see
section 6.4 of [2].

As an example, if a variant list contains the variant description

{"paper.html.en" 0.7 {type text/html} {language fr}}

and if the request contains the Accept- headers

Accept: text/html:q=1.0, */*:q=0.8
Accept-Language: en;q=1.0, fr;q=0.5

the remote variant selection algorithm will compute an overall
quality for the variant as follows:

{"paper.html.fr" 0.7 {type text/html} {language fr}}


V V V
round5 ( 0.7 * 1.0 * 0.5 ) = 0.35000

With the above Accept- headers, the complete variant list

{"paper.html.en" 0.9 {type text/html} {language en}},
{"paper.html.fr" 0.7 {type text/html} {language fr}},
{"paper.ps.en" 1.0 {type application/postscript} {language en}}

would yield the following computations:

round5 ( qs * qt * qc * ql * qf ) = Q
--- --- --- --- --- -------
paper.html.en: 0.9 * 1.0 * 1.0 * 1.0 * 1.0 = 0.90000
paper.html.fr: 0.7 * 1.0 * 1.0 * 0.5 * 1.0 = 0.35000
paper.ps.en: 1.0 * 0.8 * 1.0 * 1.0 * 1.0 = 0.80000

3.4 Definite and speculative quality values

A computed overall quality value can be either definite or
speculative. An overall quality value is definite if it was computed
without using any wildcard characters '*' in the Accept- headers, and
without the need to use the absence of a particular Accept- header.
An overall quality value is speculative otherwise.

As an example, in the previous section, the quality values of
paper.html.en and paper.html.fr are definite, and the quality value
of paper.ps.en is speculative because the type application/postscript
was matched to the range */*.

Definiteness can be defined more formally as follows. An overall
quality value Q is definite if the same quality value Q can be
computed after the request message is changed in the following way:

1. If an Accept, Accept-Charset, Accept-Language, or
Accept-Features header is missing from the request, add this
header with an empty field.

2. Delete any media ranges containing a wildcard character '*'
from the Accept header. Delete any wildcard '*' from the
Accept-Charset, Accept-Language, and Accept-Features headers.

As another example, the overall quality factor for the variant

{"blah.html" 1 {language en-gb} {features blebber [x y]}}

is 1 and definite with the Accept- headers

Accept-Language: en-gb, fr
Accept-Features: blebber, x, !y, *

and

Accept-Language: en, fr
Accept-Features: blebber, x, *

The overall quality factor is still 1, but speculative, with the
Accept- headers

Accept-language: en-gb, fr
Accept-Features: blebber, !y, *

and

Accept-Language: fr, *
Accept-Features: blebber, x, !y, *

3.5 Determining the result

The best variant, as determined by the remote variant selection
algorithm, is the one variant with the highest overall quality value,
or, if there are multiple variants which share the highest overall
quality, the first variant in the list with this value.

The end result of the remote variant selection algorithm is
Choice_response if all of the following conditions are met

a. the overall quality value of the best variant is greater
than 0

b. the overall quality value of the best variant is a definite
quality value

c. the variant resource is a neighbor of the negotiable
resource. This last condition exists to ensure that a
security-related restriction on the generation of choice
responses is met, see sections 10.2 and 14.2 of [2].

In all other cases, the end result is List_response.

The requirement for definiteness above affects the interpretation of
Accept- headers in a dramatic way. For example, it causes the remote
algorithm to interpret the header

Accept: image/gif;q=0.9, */*;q=1.0

as

`I accept image/gif with a quality of 0.9, and assign quality

factors up to 1.0 to other media types. If this information is
insufficient to make a choice on my behalf, do not make a choice
but send the list of variants'.

Without the requirement, the interpretation would have been

`I accept image/gif with a quality of 0.9, and all other media
types with a quality of 1.0'.

4 Use of the algorithm

This section discusses how user agents can use the remote algorithm
in an optimal way. This section is not normative, it is included for
informational purposes only.

4.1 Using quality factors to rank preferences

Using quality factors, a user agent can not only rank the elements
within a particular Accept- header, it can also express precedence
relations between the different Accept- headers. Consider for
example the following variant list:

{"paper.english" 1.0 {language en} {charset ISO-8859-1}},
{"paper.greek" 1.0 {language el} {charset ISO-8859-7}}

and suppose that the user prefers "el" over "en", while the user
agent can render "ISO-8859-1" with a higher quality than "ISO-8859-
7". If the Accept- headers are

Accept-Language: gr, en;q=0.8
Accept-Charset: ISO-8859-1, ISO-8859-7;q=0.6, *

then the remote variant selection algorithm would choose the English
variant, because this variant has the least overall quality
degradation. But if the Accept- headers are

Accept-Language: gr, en;q=0.8
Accept-Charset: ISO-8859-1, ISO-8859-7;q=0.95, *

then the algorithm would choose the Greek variant. In general, the
Accept- header with the biggest spread between its quality factors
gets the highest precedence. If a user agent allows the user to set
the quality factors for some headers, while other factors are hard-
coded, it should use a low spread on the hard-coded factors and a
high spread on the user-supplied factors, so that the user settings
take precedence over the built-in settings.

4.2 Construction of short requests

In a request on a transparently negotiated resource, a user agent
need not send a very long Accept- header, which lists all of its
capabilities, to get optimal results. For example, instead of
sending

Accept: image/gif;q=0.9, image/jpeg;q=0.8, image/png;q=1.0,
image/tiff;q=0.5, image/ief;q=0.5, image/x-xbitmap;q=0.8,
application/plugin1;q=1.0, application/plugin2;q=0.9

the user agent can send

Accept: image/gif;q=0.9, */*;q=1.0

It can send this short header without running the risk of getting a
choice response with, say, an inferior image/tiff variant. For
example, with the variant list

{"x.gif" 1.0 {type image/gif}}, {"x.tiff" 1.0 {type image/tiff}},

the remote algorithm will compute a definite overall quality of 0.9
for x.gif and a speculative overall quality value of 1.0 for x.tiff.
As the best variant has a speculative quality value, the algorithm
will not choose x.tiff, but return a list response, after which the
selection algorithm of the user agent will correctly choose x.gif.
The end result is the same as if the long Accept- header above had
been sent.

Thus, user agents can vary the length of the Accept- headers to get
an optimal tradeoff between the speed with which the first request is
transmitted, and the chance that the remote algorithm has enough
information to eliminate a second request.

4.2.1 Collapsing Accept- header elements

This section discusses how a long Accept- header which lists all
capabilities and preferences can be safely made shorter. The remote
variant selection algorithm is designed in such a way that it is
always safe to shorten an Accept or Accept-Charset header by two
taking two header elements `A;q=f' and `B;q=g' and replacing them by
a single element `P;q=m' where P is a wildcard pattern that matches
both A and B, and m is the maximum of f and g. Some examples are

text/html;q=1.0, text/plain;q=0.8 --> text/*;q=1.0
image/*;q=0.8, application/*;q=0.7 --> */*;q=0.8

iso-8859-5;q=1.0, unicode-1-1;q=0.8 --> *;q=1.0

Note that every `;q=1.0' above is optional, and can be omitted:

iso-8859-7;q=0.6, * --> *

For Accept-Language, it is safe to collapse all language ranges
with the same primary tag into a wildcard:

en-us;q=0.9, en-gb;q=0.7, en;q=0.8, da --> *;q=0.9, da

It is also safe to collapse a language range into a wildcard, or to
replace it by a wildcard, if its primary tag appears only once:

*;q=0.9, da --> *

Finally, in the Accept-Features header, every feature expression
can be collapsed into a wildcard, or replaced by a wildcard:

colordepth!=5, * --> *

4.2.2 Omitting Accept- headers

According to the HTTP/1.1 specification [1], the complete absence of
an Accept header from the request is equivalent to the presence of
`Accept: */*'. Thus, if the Accept header is collapsed to `Accept:
*/*', a user agent may omit it entirely. An Accept-Charset, Accept-
Language, or Accept-Features header which only contains `*' may also
be omitted.

4.2.3 Dynamically lengthening requests

In general, a user agent capable of transparent content negotiation
can send short requests by default. Some short Accept- headers could
be included for the benefit of existing servers which use HTTP/1.0
style negotiation (see section 4.2 of [2]). An example is

GET /paper HTTP/1.1
Host: x.org
User-Agent: WuxtaWeb/2.4
Negotiate: 1.0
Accept-Language: en, *;q=0.9

If the Accept- headers included in such a default request are not
suitable as input to the remote variant selection algorithm, the user
agent can disable the algorithm by sending `Negotiate: trans' instead
of `Negotiate: 1.0'.

If the user agent discovers, though the receipt of a list or choice
response, that a particular origin server contains transparently
negotiated resources, it could dynamically lengthen future requests
to this server, for example to

GET /paper/chapter1 HTTP/1.1
Host: x.org
User-Agent: WuxtaWeb/2.4
Negotiate: 1.0
Accept: text/html, application/postscript;q=0.8, */*
Accept-Language: en, fr;q=0.5, *;q=0.9
Accept-Features: tables, *

This will increase the chance that the remote variant selection
algorithm will have sufficient information to choose on behalf of the
user agent, thereby optimizing the negotiation process. A good
strategy for dynamic extension would be to extend the headers with
those media types, languages, charsets, and feature tags mentioned in
the variant lists of past responses from the server.

4.3 Differences between the local and the remote algorithm

A user agent can only optimize content negotiation though the use of
a remote algorithm if its local algorithm will generally make the
same choice. If a user agent receives a choice response containing a
variant X selected by the remote algorithm, while the local algorithm
would have selected Y, the user agent has two options:

1. Retrieve Y in a subsequent request. This is sub-optimal
because it takes time.

2. Display X anyway. This is sub-optimal because it makes the
end result of the negotiation process dependent on factors that
can randomly change. For the next request on the same resource,
and intermediate proxy cache could return a list response, which
would cause the local algorithm to choose and retrieve Y instead
of X. Compared to a stable representation, a representation
which randomly switches between X and Y (say, the version with
and without frames) has a very low subjective quality for most
users.

As both alternatives above are unattractive, a user agent should try
to avoid the above situation altogether. The sections below discuss
how this can be done.

4.3.1 Avoiding major differences

If the user agent enables the remote algorithm in this specification,
it should generally use a local algorithm which closely resembles the
remote algorithm. The algorithm should for example also use
multiplication to combine quality factors. If the user agent
combines quality factors by addition, it would be more advantageous
to define a new remote variant selection algorithm, with a new major
version number, for use by this agent.

4.3.2 Working around minor differences

Even if a local algorithm uses multiplication to combine quality
factors, it could use an extended quality formulae like

Q = round5( qs * qt * qc * ql * qf ) * q_adjust

in order to account for special interdependencies between dimensions,
which are due to limitations of the user agent. For example, if the
user agent, for some reason, cannot handle the iso-8859-7 charset
when rendering text/plain documents, the q_adjust factor would be 0
when the text/plain - iso-8859-7 combination is present in the
variant description, and 1 otherwise.

By selectively withholding information from the remote variant
selection algorithm, the user agent can ensure that the remote
algorithm will never make a choice if the local q_adjust is less than
1. For example, to prevent the remote algorithm from ever returning
a text/plain - iso-8859-7 choice response, the user agent should take
care to never produce a request which exactly specifies the quality
factors of both text/plain and iso-8859-7. The omission of either
factor from a request will cause the overall quality value of any
text/plain - iso-8859-7 variant to be speculative, and variants with
speculative quality values can never be returned in a choice
response.

In general, if the local q_adjust does not equal 1 for a particular
combination X - Y - Z, then a remote choice can be prevented by
always omitting at least one of the elements of the combination from
the Accept- headers, and adding a suitable wildcard pattern to match
the omitted element, if such a pattern is not already present.

5 Security and privacy considerations

This specification introduces no security and privacy considerations
not already covered in [2]. See [2] for a discussion of privacy
risks connected to the sending of Accept- headers.

6 Acknowledgments

Work on HTTP content negotiation has been done since at least 1993.
The authors are unable to trace the origin of many of the ideas
incorporated in this document. Many members of the HTTP working
group have contributed to the negotiation model in this
specification. The authors wish to thank the individuals who have
commented on earlier versions of this document, including Brian
Behlendorf, Daniel DuBois, Ted Hardie, Larry Masinter, and Roy T.
Fielding.

7 References

[1] Fielding, R., Gettys, J., Mogul, J., Frystyk, H., and
T. Berners-Lee, "Hypertext Transfer Protocol -- HTTP/1.1", RFC
2068, January 1997.

[2] Holtman, K., and A. Mutz, "Transparent Content Negotiation in
HTTP", RFC2295, March 1998.

8 Authors' Addresses

Koen Holtman
Technische Universiteit Eindhoven
Postbus 513
Kamer HG 6.57
5600 MB Eindhoven (The Netherlands)

EMail: koen@win.tue.nl

Andrew H. Mutz
Hewlett-Packard Company
1501 Page Mill Road 3U-3
Palo Alto CA 94304, USA

Fax: +1 415 857 4691
EMail: mutz@hpl.hp.com

9 Full Copyright Statement

Copyright (C) The Internet Society (1998). 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.