Request for Comments: 3914 Nortel Networks
Category: Informational A. Rousskov
The Measurement Factory
October 2004
Open Pluggable Edge Services (OPES) Treatment of
IAB Considerations
Status of this Memo
This memo provides information for the Internet community. It does
not specify an Internet standard of any kind. Distribution of this
memo is unlimited.
Copyright Notice
Copyright (C) The Internet Society (2004).
Abstract
IETF Internet Architecture Board (IAB) expressed nine architecture-
level considerations for the Open Pluggable Edge Services (OPES)
framework. This document describes how OPES addresses those
considerations.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 2
2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 3
3. Consideration (2.1) ’One-party consent’ . . . . . . . . . . . 3
4. Consideration (2.2) ’IP-layer communications’ . . . . . . . . 4
5. Notification Considerations . . . . . . . . . . . . . . . . . 5
5.1. Notification versus trace. . . . . . . . . . . . . . . . 6
5.2. An example of an OPES trace for HTTP . . . . . . . . . . 8
5.3. Consideration (3.1) ’Notification’ . . . . . . . . . . . 9
5.4. Consideration (3.2) ’Notification’ . . . . . . . . . . . 10
6. Consideration (3.3) ’Non-blocking’ . . . . . . . . . . . . . . 10
7. Consideration (4.1) ’URI resolution’ . . . . . . . . . . . . . 11
8. Consideration (4.2) ’Reference validity’ . . . . . . . . . . . 11
9. Consideration (4.3) ’Addressing extensions’ . . . . . . . . . 12
10. Consideration (5.1) ’Privacy’ . . . . . . . . . . . . . . . . 12
11. Consideration ’Encryption’ . . . . . . . . . . . . . . . . . . 12
12. Security Considerations . . . . . . . . . . . . . . . . . . . 13
13. Compliance . . . . . . . . . . . . . . . . . . . . . . . . . . 13
14. References . . . . . . . . . . . . . . . . . . . . . . . . . . 14
14.1. Normative References . . . . . . . . . . . . . . . . . . 14
14.2. Informative References . . . . . . . . . . . . . . . . . 14
Authors’ Addresses . . . . . . . . . . . . . . . . . . . . . . . . 15
Full Copyright Statement . . . . . . . . . . . . . . . . . . . . . 16
1. Introduction
The Open Pluggable Edge Services (OPES) architecture [RFC3835],
enables cooperative application services (OPES services) between a
data provider, a data consumer, and zero or more OPES processors.
The application services under consideration analyze and possibly
transform application-level messages exchanged between the data
provider and the data consumer.
In the process of chartering OPES, the IAB made recommendations on
issues that OPES solutions should be required to address. These
recommendations were formulated in the form of a specific IAB
considerations document [RFC3238]. In that document, IAB emphasized
that its considerations did not recommend specific solutions and did
not mandate specific functional requirements. Addressing an IAB
consideration may involve showing appropriate protocol mechanisms or
demonstrating that the issue does not apply. Addressing a
consideration does not necessarily mean supporting technology implied
by the consideration wording.
The primary goal of this document is to show that all formal IAB
recommendations are addressed by OPES, to the extent that those
considerations can be addressed by an IETF working group. The
limitations of OPES working group to address certain aspects of IAB
considerations are also explicitly documented.
IAB considerations document [RFC3238] contains many informal
recommendations. For example, while the IAB informally requires OPES
architecture to "protect end-to-end data integrity by supporting
end-host detection and response to inappropriate behavior by OPES
intermediaries", the IAB has chosen to formalize these requirements
via a set of more specific recommendations, such as Notification
considerations addressed in Section 5.3 and Section 5.4 below. OPES
framework addresses informal IAB recommendations by addressing
corresponding formal considerations.
There are nine formal IAB considerations [RFC3238] that OPES has to
address. In the core of this document are the corresponding nine
"Consideration" sections. For each IAB consideration, its section
contains general discussion as well as references to specific OPES
mechanisms relevant to the consideration.
2. Terminology
This document does not introduce any new terminology but uses
terminology from other OPES documents.
3. Consideration (2.1) ’One-party consent’
"An OPES framework standardized in the IETF must require that the use
of any OPES service be explicitly authorized by one of the
application-layer end-hosts (that is, either the content provider or
the client)." [RFC3238]
OPES architecture requires that "OPES processors MUST be consented to
by either the data consumer or data provider application" [RFC3835].
While this requirement directly satisfies IAB concern, no requirement
alone can prevent consent-less introduction of OPES processors. In
other words, OPES framework requires one-party consent but cannot
guarantee it in the presence of incompliant OPES entities.
In [RFC3897], the OPES architecture enables concerned parties to
detect unwanted OPES processors by examining OPES traces. While the
use of traces in OPES is mandatory, a tracing mechanism on its own
cannot detect processors that are in violation of OPES
specifications. Examples include OPES processors operating in
stealth mode. However, the OPES architecture allows the use of
content signature to verify the authenticity of performed
adaptations. Content signatures is a strong but expensive mechanism
that can detect any modifications of signed content provided that the
content provider is willing to sign the data and that the client is
willing to either check the signature or relay received content to
the content provider for signature verification.
OPES entities may copy or otherwise access content without modifying
it. Such access cannot be detected using content signatures. Thus,
"passive" OPES entities can operate on signed content without the
data consumer or provider consent. If content privacy is a concern,
then content encryption can be used. A passive processor is no
different from any intermediary operating outside of OPES framework.
No OPES mechanism (existing or foreseeable) can prevent non-modifying
access to content.
In summary, the one-party consent is satisfied by including the
corresponding requirement in the IAB architecture document. That
requirement alone cannot stop incompliant OPES entities to perform
consent-less adaptations, but OPES framework allows for various means
of detecting and/or preventing such adaptations. These means
typically introduce overheads and require some level of producer-
consumer cooperation.
4. Consideration (2.2) ’IP-layer communications’
"For an OPES framework standardized in the IETF, the OPES
intermediary must be explicitly addressed at the IP layer by the end
user" [RFC3238].
The OPES architecture requires that "OPES processors MUST be
addressable at the IP layer by the end user (data consumer
application)" [RFC3835]. The IAB and the architecture documents
mention an important exception: addressing the first OPES processor
in a chain of processors is sufficient. That is, a chain of OPES
processors is viewed as a single OPES "system" at the address of the
first chain element.
The notion of a chain is not strictly defined by IAB. For the
purpose of addressing this consideration, a group of OPES processors
working on a given application transaction is considered. Such a
group would necessarily form a single processing chain, with a single
"exit" OPES processor (i.e., the processor that adapted the given
message last). The OPES architecture essentially requires that last
OPES processor to be explicitly addressable at the IP layer by the
data consumer application. The chain formation, including its exit
point may depend on an application message and other dynamic factors
such as time of the day or system load.
Furthermore, if OPES processing is an internal processing step at a
data consumer or a data provider application side, then the last OPES
processor may reside in a private address space and may not be
explicitly addressable from the outside. In such situations, the
processing side must designate an addressable point on the same
processing chain. That designated point may not be, strictly
speaking, an OPES processor, but it will suffice as such as far as
IAB considerations are concerned -- the data consumer application
will be able to address it explicitly at the IP layer and it will
represent the OPES processing chain to the outside world.
Designating an addressable processing point avoids the conflict
between narrow interpretation of the IAB consideration and real
system designs. It is irrational to expect a content provider to
provide access to internal hosts participating in content generation,
whether OPES processors are involved or not. Moreover, providing
such access would serve little practical purpose because internal
OPES processors are not likely to be able to answer any data consumer
queries, being completely out of content generation context. For
example, an OPES processor adding customer-specific information to
XML pages may not understand or be aware of any final HTML content
that the data consumer application receives and may not be able to
map end user request to any internal user identification. Since OPES
requires the end of the message processing chain to be addressable,
the conflict does not exist. OPES places no requirements on the
internal architecture of data producer systems while requiring the
entire OPES-related content production "system" to be addressable at
the IP layer.
Private Domain | Public Domain | Private Domain
| |
+--------------+ | +-------------+ +--------+
| Data | | | OPES System | |Data |
| Consumer |<--- network -->| with public |<---->|Provider|
| Application | | | IP address | |App |
+--------------+ | +-------------+ +--------+
| |
| |
Figure 1
5. Notification Considerations
This section discusses how OPES framework addresses IAB Notification
considerations 3.1 and 3.2.
5.1. Notification versus trace
Before specific considerations are discussed, the relationship
between IAB notifications and OPES tracing has to be explained. OPES
framework concentrates on tracing rather than notification. The OPES
Communications specification [RFC3897] defines "OPES trace" as
application message information about OPES entities that adapted the
message. Thus, OPES trace follows the application message it traces.
The trace is for the recipient of the application message. Traces
are implemented as extensions of application protocols being adapted
and traced.
As opposed to an OPES trace, provider notification (as implied by
IAB) notifies the sender of the application message rather than the
recipient. Thus, notifications propagate in the opposite direction
of traces. Supporting notifications directly would require a new
protocol. Figure 2 illustrates the differences between a trace and
notification from a single application message point of view.
sender --[message A]--> OPES --[message A’]--> recipient
^ V [with trace]
| |
+-<-- [notification] ---+
Figure 2
Since notifications cannot be piggy-backed to application messages,
they create new messages and may double the number of messages the
sender has to process. The number of messages that need to be
proceed is larger if several intermediaries on the message path
generate notifications. Associating notifications with application
messages may require duplicating application message information in
notifications and may require maintaining a sender state until
notification is received. These actions increase the performance
overhead of notifications.
The level of available details in notifications versus provider
interest in supporting notification is another concern. Experience
shows that content providers often require very detailed information
about user actions to be interested in notifications at all. For
example, Hit Metering protocol [RFC2227] has been designed to supply
content providers with proxy cache hit counts, in an effort to reduce
cache busting behavior which was caused by content providers desire
to get accurate site "access counts". However, the Hit Metering
protocol is currently not widely deployed because the protocol does
not supply content providers with information such as client IP
addresses, browser versions, or cookies.
Hit Metering experience is relevant because Hit Metering protocol was
designed to do for HTTP caching intermediaries what OPES
notifications are meant to do for OPES intermediaries. Performance
requirements call for state reduction via aggregation of
notifications while provider preferences call for state preservation
or duplication. Achieving the right balance when two sides belong to
different organizations and have different optimization priorities
may be impossible.
Thus, instead of explicitly supporting notifications at the protocol
level, OPES concentrates on tracing facilities. In essence, OPES
supports notifications indirectly, using tracing facilities. In
other words, the IAB choice of "Notification" label is interpreted as
"Notification assistance" (i.e., making notifications meaningful) and
is not interpreted as a "Notification protocol".
The above concerns call for making notification optional. The OPES
architecture allows for an efficient and meaningful notification
protocol to be implemented in certain OPES environments. For
example, an OPES callout server attached to a gateway or firewall may
scan outgoing traffic for signs of worm or virus activity and notify
a local Intrusion Detection System (IDS) of potentially compromised
hosts (e.g., servers or client PCs) inside the network. Such
notifications may use OPES tracing information to pinpoint the
infected host (which could be another OPES entity). In this example,
notifications are essentially sent back to the content producer (the
local network) and use OPES tracing to supply details.
Another environment where efficient and meaningful notification using
OPES tracing is possible are Content Delivery Networks (CDNs). A CDN
node may use multiple content adaptation services to customize
generic content supplied by the content producer (a web site). For
example, a callout service may insert advertisements for client-local
events. The CDN node itself may not understand specifics of the ad
insertion algorithm implemented at callout servers. However, the
node may use information in the OPES trace (e.g., coming from the
callout service) to notify the content producer. Such notifications
may be about the number of certain advertisements inserted (i.e., the
number of "impressions" delivered to the customer) or even the number
of ad "clicks" the customer made (e.g., if the node hosts content
linked from the ads). Callout services doing ad insertion may lack
details (e.g., a customer ID/address or a web server authentication
token) to contact the content producer directly in this case. Thus,
OPES trace produced by an OPES service becomes essential in enabling
meaningful notifications that the CDN node sends to the content
producer.
5.2. An example of an OPES trace for HTTP
The example below illustrates adaptations done to HTTP request at an
OPES processor operated by the client ISP. Both original (as sent by
an end user) and adapted (as received by the origin web server)
requests are shown. The primary adaptation is the modification of
HTTP "Accept" header. The secondary adaptation is the addition of an
OPES-System HTTP extension header [I-D.ietf-opes-http].
GET /pub/WWW/ HTTP/1.1
Host: www.w3.org
Accept: text/plain
Figure 3
... may be adapted by an ISP OPES system to become:
GET /pub/WWW/ HTTP/1.1
Host: www.w3.org
Accept: text/plain; q=0.5, text/html, text/x-dvi; q=0.8
OPES-System: http://www.isp-example.com/opes/?client-hash=1234567
Figure 4
The example below illustrates adaptations done to HTTP response at an
OPES intermediary operated by a Content Distribution Network (CDN).
Both original (as sent by the origin web server) and adapted (as
received by the end user) responses are shown. The primary
adaptation is the conversion from HTML markup to plain text. The
secondary adaptation is the addition of an OPES-System HTTP extension
header.
HTTP/1.1 200 OK
Content-Length: 12345
Content-Encoding: text/html
<html><head><h1>Available Documenta...
Figure 5
... may be adapted by a CDN OPES system to become:
HTTP/1.1 200 OK
Content-Length: 2345
Content-Encoding: text/plain
OPES-System: http://www.cdn-example.com/opes/?site=7654&svc=h2t
AVAILABLE DOCUMENTA...
Figure 6
In the above examples, OPES-System header values contain URIs that
may point to OPES-specific documents such as description of the OPES
operator and its privacy policy. Those documents may be
parameterized to allow for customizations specific to the transaction
being traced (e.g., client or even transaction identifier may be used
to provide more information about performed adaptations). An OPES-
Via header may be used to provide a more detailed trace of specific
OPES entities within an OPES System that adapted the message. Traced
OPES URIs may be later used to request OPES bypass [RFC3897].
5.3. Consideration (3.1) ’Notification’
"The overall OPES framework needs to assist content providers in
detecting and responding to client-centric actions by OPES
intermediaries that are deemed inappropriate by the content provider"
[RFC3238].
OPES tracing mechanisms assist content providers in detecting
client-centric actions by OPES intermediaries. Specifically, a
compliant OPES intermediary or system notifies a content provider of
its presence by including its tracing information in the application
protocol requests. An OPES system MUST leave its trace [RFC3897].
Detection assistance has its limitations. Some OPES intermediaries
may work exclusively on responses and may not have a chance to trace
the request. Moreover, some application protocols may not have
explicit requests (e.g., a content push service).
OPES tracing mechanisms assist content providers in responding to
client-centric actions by OPES intermediaries. Specifically, OPES
traces MUST include identification of OPES systems and SHOULD include
a list of adaptation actions performed on provider’s content. This
tracing information may be included in the application request.
Usually, however, this information will be included in the
application response, an adapted version of which does not reach the
content provider. If OPES end points cooperate, then notification
can be assisted with traces. Content providers that suspect or
experience difficulties can do any of the following:
o Check whether requests they receive pass through OPES
intermediaries. Presence of OPES tracing info will determine
that. This check is only possible for request/response protocols.
For other protocols (e.g., broadcast or push), the provider would
have to assume that OPES intermediaries are involved until proven
otherwise.
o If OPES intermediaries are suspected, request OPES traces from
potentially affected user(s). The trace will be a part of the
application message received by the user software. If involved
parties cooperate, the provider(s) may have access to all the
needed information. Certainly, lack of cooperation may hinder
access to tracing information. To encourage cooperation, data
providers might be able to deny service to uncooperative users.
o Some traces may indicate that more information is available by
accessing certain resources on the specified OPES intermediary or
elsewhere. Content providers may query for more information in
this case.
o If everything else fails, providers can enforce no-adaptation
policy using appropriate OPES bypass mechanisms and/or end-to-end
encryption mechanisms.
OPES detection and response assistance is limited to application
protocols with support for tracing extensions. For example, HTTP
[RFC2616] has such support while DNS over UDP does not.
5.4. Consideration (3.2) ’Notification’
"The overall OPES framework should assist end users in detecting the
behavior of OPES intermediaries, potentially allowing them to
identify imperfect or compromised intermediaries" [RFC3238].
OPES tracing mechanisms assist end users in detecting OPES
intermediaries. Specifically, a compliant OPES intermediary or
system notifies an end user of its presence by including its tracing
information in the application protocol messages sent to the client.
An OPES system MUST leave its trace [RFC3897]. However, detection
assistance has its limitations. Some OPES systems may work
exclusively on requests and may not have a chance to trace the
response. Moreover, some application protocols may not have explicit
responses (e.g., event logging service).
OPES detection assistance is limited to application protocols with
support for tracing extensions. For example, HTTP [RFC2616] has such
support while DNS over UDP does not.
6. Consideration (3.3) ’Non-blocking’
"If there exists a "non-OPES" version of content available from the
content provider, the OPES architecture must not prevent users from
retrieving this "non-OPES" version from the content provider"
[RFC3238].
"OPES entities MUST support a bypass feature" [RFC3897]. If an
application message includes bypass instructions and an OPES
intermediary is not configured to ignore them, the matching OPES
intermediary will not process the message. An OPES intermediary may
be configured to ignore bypass instructions only if no non-OPES
version of content is available. Bypass may generate content errors
since some OPES services may be essential but may not be configured