One document matched: draft-ietf-rmt-sec-discussion-04.txt
Differences from draft-ietf-rmt-sec-discussion-03.txt
RMT B. Adamson
Internet-Draft Naval Research Laboratory
Intended status: Informational V. Roca
Expires: April 22, 2010 INRIA
H. Asaeda
Keio University
October 19, 2009
Security and Reliable Multicast Transport Protocols: Discussions and
Guidelines
draft-ietf-rmt-sec-discussion-04
Status of this Memo
This Internet-Draft is submitted to IETF in full conformance with the
provisions of BCP 78 and BCP 79. This document may contain material
from IETF Documents or IETF Contributions published or made publicly
available before November 10, 2008. The person(s) controlling the
copyright in some of this material may not have granted the IETF
Trust the right to allow modifications of such material outside the
IETF Standards Process. Without obtaining an adequate license from
the person(s) controlling the copyright in such materials, this
document may not be modified outside the IETF Standards Process, and
derivative works of it may not be created outside the IETF Standards
Process, except to format it for publication as an RFC or to
translate it into languages other than English.
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-
Drafts.
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."
The list of current Internet-Drafts can be accessed at
http://www.ietf.org/ietf/1id-abstracts.txt.
The list of Internet-Draft Shadow Directories can be accessed at
http://www.ietf.org/shadow.html.
This Internet-Draft will expire on April 22, 2010.
Copyright Notice
Adamson, et al. Expires April 22, 2010 [Page 1]
Internet-Draft Security and RMT Protocols October 2009
Copyright (c) 2009 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents in effect on the date of
publication of this document (http://trustee.ietf.org/license-info).
Please review these documents carefully, as they describe your rights
and restrictions with respect to this document.
Abstract
This document describes general security considerations for the
Reliable Multicast Transport (RMT) Working Group set of building
blocks and protocols. An emphasis is placed on risks that might be
resolved in the scope of transport protocol design. However,
relevant security issues related to IP Multicast control-plane and
other concerns not strictly within the scope of reliable transport
protocol design are also discussed. The document also begins an
exploration of approaches that could be embraced to mitigate these
risks. The purpose of this document is to provide a consolidated
security discussion and provide a basis for further discussions and
potential resolution of any significant security issues that may
exist in the current set of RMT standards.
Adamson, et al. Expires April 22, 2010 [Page 2]
Internet-Draft Security and RMT Protocols October 2009
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1. Conventions Used in this Document . . . . . . . . . . . . 6
2. Quick Introduction to RMT Protocols and their Use . . . . . . 6
2.1. The Two Families of CDP . . . . . . . . . . . . . . . . . 6
2.2. RMT Protocol Characteristics . . . . . . . . . . . . . . . 7
2.3. Target Use Case Characteristics . . . . . . . . . . . . . 7
3. Some Security Threats . . . . . . . . . . . . . . . . . . . . 8
3.1. Control-Plane Attacks . . . . . . . . . . . . . . . . . . 9
3.1.1. Control Plane Monitoring . . . . . . . . . . . . . . . 9
3.1.2. Unauthorized (or Malicious) Group Membership . . . . . 10
3.2. Data-Plane Attacks . . . . . . . . . . . . . . . . . . . . 10
3.2.1. Rogue Traffic Generation . . . . . . . . . . . . . . . 11
3.2.2. Sender Message Spoofing . . . . . . . . . . . . . . . 11
3.2.3. Receiver Message Spoofing . . . . . . . . . . . . . . 12
3.2.4. Replay Attacks . . . . . . . . . . . . . . . . . . . . 12
4. General Security Goals . . . . . . . . . . . . . . . . . . . . 13
4.1. Network Protection . . . . . . . . . . . . . . . . . . . . 14
4.2. Protocol Protection . . . . . . . . . . . . . . . . . . . 14
4.3. Content Protection . . . . . . . . . . . . . . . . . . . . 14
4.4. Privacy . . . . . . . . . . . . . . . . . . . . . . . . . 15
5. Elementary Security Techniques . . . . . . . . . . . . . . . . 15
6. Technological Building Blocks . . . . . . . . . . . . . . . . 17
6.1. IPsec . . . . . . . . . . . . . . . . . . . . . . . . . . 17
6.1.1. Benefits . . . . . . . . . . . . . . . . . . . . . . . 17
6.1.2. Requirements . . . . . . . . . . . . . . . . . . . . . 18
6.1.3. Limitations . . . . . . . . . . . . . . . . . . . . . 18
6.2. Group MAC . . . . . . . . . . . . . . . . . . . . . . . . 19
6.2.1. Benefits . . . . . . . . . . . . . . . . . . . . . . . 19
6.2.2. Requirements . . . . . . . . . . . . . . . . . . . . . 19
6.2.3. Limitations . . . . . . . . . . . . . . . . . . . . . 19
6.3. Digital Signatures . . . . . . . . . . . . . . . . . . . . 19
6.3.1. Benefits . . . . . . . . . . . . . . . . . . . . . . . 19
6.3.2. Requirements . . . . . . . . . . . . . . . . . . . . . 20
6.3.3. Limitations . . . . . . . . . . . . . . . . . . . . . 20
6.4. TESLA . . . . . . . . . . . . . . . . . . . . . . . . . . 20
6.4.1. Benefits . . . . . . . . . . . . . . . . . . . . . . . 20
6.4.2. Requirements . . . . . . . . . . . . . . . . . . . . . 21
6.4.3. Limitations . . . . . . . . . . . . . . . . . . . . . 21
6.5. Source-Specific Multicast . . . . . . . . . . . . . . . . 21
6.5.1. Requirements . . . . . . . . . . . . . . . . . . . . . 22
6.5.2. Limitations . . . . . . . . . . . . . . . . . . . . . 22
6.5.3. Source-Based and Receiver-Based Attacks . . . . . . . 22
6.6. Summary . . . . . . . . . . . . . . . . . . . . . . . . . 23
7. Security Infrastructure . . . . . . . . . . . . . . . . . . . 23
8. New Threats Introduced by the Security Scheme Itself . . . . . 24
9. Consequences for the RMT and MSEC Working Group . . . . . . . 24
Adamson, et al. Expires April 22, 2010 [Page 3]
Internet-Draft Security and RMT Protocols October 2009
9.1. RMT Transport Message Security Encapsulation Header . . . 24
10. Security Considerations . . . . . . . . . . . . . . . . . . . 25
11. Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . 25
12. References . . . . . . . . . . . . . . . . . . . . . . . . . . 25
12.1. Normative References . . . . . . . . . . . . . . . . . . . 25
12.2. Informative References . . . . . . . . . . . . . . . . . . 27
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 27
Adamson, et al. Expires April 22, 2010 [Page 4]
Internet-Draft Security and RMT Protocols October 2009
1. Introduction
The Reliable Multicast Transport (RMT) Working Group has produced a
set of building block (BB) and protocol instantiation (PI)
specifications for reliable multicast data transport. Some present
PIs defined within the scope of RMT include Asynchronous Layered
Coding (ALC) [I-D.ietf-rmt-pi-alc-revised], NACK-Oriented Reliable
Multicast (NORM) [I-D.ietf-rmt-pi-norm-revised], and the File
Delivery over Unidirectional Transport (FLUTE)
[I-D.ietf-rmt-flute-revised] application that is built on top of ALC.
These can be considered "Content Delivery Protocols" (CDP) as
described in [Neumann05]. In this document, the term CDP will refer
indifferently to either ALC or NORM, with their associated BBs.
The use of these BBs and PIs raises some new security risks. For
instance, these protocols share a novel set of Forward Error
Correction (FEC) and congestion control building blocks that present
some new capabilities for Internet transport, but may also pose some
new security risks. Yet some security risks are not related to the
particular BBs used by the PIs, but are more general. Reliable
multicast transport sessions are expected to involve at least one
sender and multiple receivers. Thus, the risk of and avenues to
attack are implicitly greater than that of point-to-point (unicast)
transport sessions. Also the nature of IP multicast can expose other
coexistent network flows and services to risk if malicious users
exploit it. The classic Any-Source Multicast (ASM) [RFC1112] model
of multicast routing allows any host to join an IP multicast group
and send traffic to that group. This poses many potential security
challenges. And, while the emerging Source-Specific Multicast (SSM)
[RFC3569], [RFC4607] model that enables users to receive multicast
data sent only from specified sender(s) simplifies some challenges,
there are still specific issues. For instance, possible areas of
attack include those against the control plane where malicious hosts
join IP multicast groups to cause multicast traffic to be directed to
parts of the network where it is not needed or desired. This may
indirectly cause denial-of-service (DoS) to other network flows.
Also, attackers may transmit erroneous or corrupt messages to the
group or employ strategies such as replay attack within the "data
plane" of protocol operation.
The goals of this document are therefore to:
1. Define the possible general security goals: protecting the
network infrastructure, and/or the protocol, and/or the content,
and/or the user (e.g., its privacy);
2. List the possible elementary security services that will make it
possible to fulfill the general security goals. Some of these
Adamson, et al. Expires April 22, 2010 [Page 5]
Internet-Draft Security and RMT Protocols October 2009
services are generic (e.g., object and/or packet integrity),
while others are specific to RMT protocols (e.g., congestion
control specific security schemes);
3. List some technological building blocks and solutions that can
provide the desired security services;
4. Highlight the CDP and the use-case specificities that will impact
security. Indeed, the set of solutions proposed to fulfill the
security goals will greatly be impacted by these considerations;
In some cases, the existing RMT documents already discuss the risks
and outline approaches to solve them, at least partially. The
purpose of this document is to consolidate this content and provide a
basis for further discussion and potential resolution of any
significant security issues that may exist.
1.1. Conventions Used in this Document
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in [RFC2119].
2. Quick Introduction to RMT Protocols and their Use
2.1. The Two Families of CDP
The ALC and NORM classes of CDP are designed to reliably deliver
content to a group of multicast receivers. However, ALC and NORM
have a different set of features and limitations. ALC supports a
unidirectional delivery model where there is no feedback from the
receivers to senders. Reliability is achieved by the joint use of
carousel-based transmission techniques associated to FEC encoding to
recover from missing (erased) packets.
On the opposite, NORM achieves reliability by means of FEC encoding
(as with ALC) and feedback from the receivers. More specifically,
NORM leverages Negative Acknowledgement techniques to control the
senders' transmission of content. The advantage is that the sender
need not transmit any more information than necessary to satisfy the
receivers' need for fully reliable transfers. However, while NORM
specifies feedback control techniques to allow it to scale to large
group sizes, it is not as massively scalable as ALC. Additionally,
the NORM feedback control mechanisms add some header content and
protocol implementation complexity.
The appropriate choice of a CDP depends upon application needs,
Adamson, et al. Expires April 22, 2010 [Page 6]
Internet-Draft Security and RMT Protocols October 2009
deployment constraints, and network connectivity considerations. And
while there are many common security considerations for these two
classes of CDP, there are also some unique considerations for each.
2.2. RMT Protocol Characteristics
This section focuses on the RMT protocol characteristics that impact
the choice of the technological building blocks, and the way they can
be applied. Both ALC and NORM have been designed with receiver group
size scalability. While ALC targets massively scalable sessions
(e.g., millions of receivers), NORM is less ambitious, essentially
because of the use of feedback messages.
The ALC and NORM protocols also differ in the communication paths:
o sender to receivers: ALC and NORM, for bulk data transfer and
signaling messages;
o receivers to sender: NORM only, for feedback messages;
o receivers to receivers: NORM only for control messages;
Note that the fact ALC is capable of working on top of purely
unidirectional networks does not mean that no back-channel is
available (Section 2.3).
The NORM and ALC protocols support a variety of content delivery
models where transport may be carefully coordinated among the sender
and receivers or with looser coordination and interaction. This
leads to a number of different use cases for these protocols.
2.3. Target Use Case Characteristics
This section focuses on the target use cases and their special
characteristics. These details will impact both the choice of the
technological building blocks and the way they can be applied. One
can distinguish the following use case features:
o Purely unidirectional transport versus symmetric bidirectional
transport versus asymmetric bidirectional transport. Most of the
time, the amount of traffic flowing to the source is limited, and
one can overlook whether the transport channel is symmetric or
not. The nature of the underlying transport channel is of
paramount importance, since many security building blocks will
require a bidirectional communication;
o Massively scalable versus moderately scalable session. Here we do
not define precisely what the terms "massively scalable" and
Adamson, et al. Expires April 22, 2010 [Page 7]
Internet-Draft Security and RMT Protocols October 2009
"moderately scalable" mean.
o Known set of receivers versus unknown set of receivers: I.e., does
the source know at any point of time the set of receivers or not?
Of course, knowing the set of receivers is usually not compatible
with massively scalable sessions;
o Dynamic set of receivers versus fixed set of receivers: I.e., does
the source know at some point of time the maximum set of receivers
or will it evolve dynamically?
o High rate data flow versus small rate data flow: Some security
building blocks are CPU-intensive and are therefore incompatible
with high data rate sessions (e.g., solutions that digitally sign
all packets sent).
o Protocol stack available at both ends: A solution that requires
some unusual features within the protocol stack will not always be
usable. Some target environments (e.g., embedded systems) provide
a minimum set of features and extending them (e.g., to add IPsec)
is not necessarily realistic;
o Multicast routing and other layer-3 protocols in use: E.g., SSM
routing is often seen as one of the key service to improve the
security within multicast sessions, and some security building
blocks require specialized versions of layer-3 protocols (e.g.,
IGMP/MLD with security extensions). In some cases these
assumptions might not be realistic.
Depending on the target goal and the associated security building
block used, other features might be of importance. For instance
TESLA requires a loose time synchronization between the source and
the receivers. Several possible techniques are available to provide
this, but some of them may be feasible only if the target use case
has the appropriate characteristics.
3. Some Security Threats
The IP architecture provides common access to notional control and
data planes to both end and intermediate systems. For the purposes
of discussion here, the "control plane" mechanisms are considered
those with message exchanges between end systems (typically
computers) and intermediate systems (typically routers) (or among
intermediate systems) while the "data plane" encompasses messages
exchanged among end systems, usually pertaining to the transfer of
application data. The security threats described here are introduced
within the taxonomy of control plane and data plane IP mechanisms.
Adamson, et al. Expires April 22, 2010 [Page 8]
Internet-Draft Security and RMT Protocols October 2009
3.1. Control-Plane Attacks
In this discussion, "control-plane" in the context of Internet
Protocol systems refers to signaling among end systems and
intermediate systems to facilitate routing and forwarding of packets.
For IP multicast, this notably includes Internet Group Management
Protocol (IGMP), Multicast Listener Discovery protocol (MLD), and
multicast routing protocol messaging. While control-plane attacks
may be considered outside of the scope of the transport protocol
specifications discussed here, it is important to understand the
potential impact of such attacks with respect to the deployment and
operation of these protocols. For example, awareness of possible IP
Multicast control-plane manipulation that can lead to unauthorized
(or unexpected) monitoring of data plane traffic by malicious users
may lead a transport application or protocol implementation to
support encryption to ensure data confidentiality and/or privacy.
Also, these types of attack also have bearing on assessing the real
risks of potentially more complex attacks against the transport
mechanisms themselves. In some cases, the solutions to these
control-plane risk areas may reduce the impact or possibility of some
data-plane attacks that are discussed in this document.
The presence of these types of attack may necessitate that policy-
based controls be embedded in routers to limit the distribution
(including transmission and reception) of multicast traffic (on a
group-wise and/or traffic volume basis) to different parts of the
network. Such policy-based controls are beyond the scope of the RMT
protocol specifications. However, such network protection mechanisms
may reduce the opportunities for or effectiveness of some of the
data-plane attacks discussed later. For example, reverse-path checks
can significantly limit opportunities for attackers to conduct replay
attacks when hosts actually do use IPsec. Also, future IP Multicast
control protocols may wish to consider providing security mechanism
to prevent unauthorized monitoring or manipulation of messages
related to group membership, routing, and activity. The sections
below describe some variants of control-plane attacks.
3.1.1. Control Plane Monitoring
While this may not be a direct attack on the transport system, it may
be possible for an attacker to gain useful information in advancing
attack goals by monitoring IP Multicast control plane traffic
including group membership and multicast routing information.
Identification of hosts and/or routers participating in specific
multicast groups may readily identify systems vulnerable to protocol-
specific exploitation. And, with regards to user privacy concerns,
such "side information" may be relevant to this emerging aspect of
network security as described in Section 4.4.
Adamson, et al. Expires April 22, 2010 [Page 9]
Internet-Draft Security and RMT Protocols October 2009
3.1.2. Unauthorized (or Malicious) Group Membership
One of the simplest attacks is that where a malicious host joins an
IP multicast group so that potentially unwanted traffic is routed to
the host's network interface. This type of attack can turn a
legitimate source of IP traffic into a "attacker" without requiring
any access privileges to the source host or routers involved. This
type of attack can be used for denial-of-service purposes or for the
real attacker (the malicious joiner) to gain access to the
information content being sent. Similarly, some routing protocols
may permit any sender (whether joined to the specific group or not)
to transmit messages to a multicast group.
It is possible that malicious hosts could also spoof IGMP/MLD
messages, joining groups posing as legitimate hosts (or spoof source
traffic from legitimate hosts). This may be done at intermediate
locations in the network or by hosts co-resident with the authorized
hosts on local area networks. Such spoofing could be done by raw
packet generation or with replay of previously-recorded control
messages.
For the sake of completeness, it should be noted that multicast
routing protocol control messaging may be subject to similar threats
if sufficient protocol security mechanisms are not enabled in the
routing infrastructure. [RFC4609] describes security threats to the
PIM-SM multicast routing infrastructures.
3.2. Data-Plane Attacks
This section discusses some types of active attacks that might be
conducted "in-band" with respect to the reliable multicast transport
protocol operating within the data plane of network data transfer.
I.e., the "data-plane" here refers to IP packets containing end-to-
end transport content to support the reliable multicast transfer.
The passive attack of unauthorized data-plan monitoring is discussed
above since such activity might be made possible by the
vulnerabilities of the IP Multicast control plane. To cover the two
classes of RMT protocols, the active data-plane attacks are
categorized as 1) those where the attacker generates messages posing
as a data sender, and 2) those where the attacker generates messages
posing as a receiver providing feedback to the sender(s) or group.
Additionally, a common threat to protocol operation is that of brute-
force, rogue packet generation. This is discussed briefly below, but
the more subtle attacks that might be conducted are given more
attention as those fall within the scope of the RMT transport
protocol design. Additionally, special consideration is given to
that of the "replay attack" [see Section 3.2.4], as it can be applied
across these different categories.
Adamson, et al. Expires April 22, 2010 [Page 10]
Internet-Draft Security and RMT Protocols October 2009
3.2.1. Rogue Traffic Generation
If an attacker is able to successfully inject packets into the
multicast distribution tree, one obvious denial-of-service attack is
for the attacker to generate a large volume of apparently
authenticate traffic (and if authentication mechanisms are used, a
"replay" attack strategy might be used). The impact of this type of
attack can be significant since the potential for routers to relay
the traffic to multiple portions of a networks (as compared to a
single unicast routing path). However, other than the amplified
negative impact to the network, this type of attack is no different
than what is possible with rogue unicast packet generation and
similar measures used to protect the network from such attacks could
be used to contain this type of brute-force attack. Of course, the
pragmatic question of whether current implementations of such
protection mechanisms support IP Multicast SHOULD be considered.
3.2.2. Sender Message Spoofing
Sender message spoofing attacks are applicable to both CDP: ALC
(sender-only transmission) and NORM (sender-receiver exchanges).
Without an authentication mechanism, an attacker can easily generate
sender messages that could disrupt a reliable multicast transfer
session. And with FEC-based transport mechanisms, a single packet
with an apparently-correct FEC payload identifier[RFC5052] but a
corrupted FEC payload could potentially render an entire block of
transported data invalid. Thus, a modest injection rate of corrupt
traffic could cause severe impairment of data transport.
Additionally, such invalid sender packets could convey out-of-bound
indices (e.g., bad symbol or block identifiers) that can lead to
buffer overflow exploits or similar issues in implementations that
insufficiently check for invalid data.
An indirect use of sender message spoofing would be to generate
messages that would cause receivers to take inappropriate congestion-
control action. In the case of the layered congestion control
mechanisms proposed for ALC use, this could lead to the receivers
erroneously leaving groups associated with higher bandwidth transport
layers and suffering unnecessarily low transport rates. Similarly,
receivers may be misled to join inappropriate groups directing
unwanted traffic to their part of the network. Attacks with similar
effect could be conducted against the TCP-Friendly Multicast
Congestion Control (TFMCC) [RFC4654] approach proposed for NORM
operation with spoofing of sender messages carrying congestion
control state to receivers.
Adamson, et al. Expires April 22, 2010 [Page 11]
Internet-Draft Security and RMT Protocols October 2009
3.2.3. Receiver Message Spoofing
These attacks are limited to CDP that use feedback from receivers in
the group to influence sender and other receiver operation. In the
NORM protocol, this includes negative-acknowledgement (NACK) messages
fed back to the sender to achieve reliable transfer, congestion
control feedback content, and the optional positive acknowledgement
features of the specification. It is also important to note that for
ASM operation, NORM receivers pay attention to the messages of other
receivers for the purpose of suppression to avoid feedback implosion
as group size grows large.
An attacker that can generate false feedback can manipulate the NORM
sender to unnecessarily transmit repair information and reduce the
goodput of the reliable transfer regardless of the sender's transmit
rate. Contrived congestion control feedback could also cause the
sender to transmit at an unfairly low rate.
As mentioned, spoofed receiver messaging may not be directed only at
senders, but also at receivers participating in the session. For
example, an attacker may direct phony receiver feedback messages to
selected receivers in the group causing those receivers to suppress
feedback that might have otherwise been transmitted. This attack
could compromise the ability of those receivers to achieve reliable
transfer. Also, suppressed congestion control feedback could cause
the sender to transmit at a rate unfair to those attacked receivers
if their fair congestion control rate were lower.
3.2.4. Replay Attacks
The infamous "replay attack" (injection of a previously transmitted
packet to one or more participants) is given special attention here
because of the special consequences it can have on RMT protocol
operation. Without specific protection mechanisms against replay
(e.g., duplicate message detection), it is possible for these attacks
to be successful even when security mechanisms such as packet
authentication and/or encryption are employed.
3.2.4.1. Replay of Sender Messages
Generally, replay of recent protocol messages from the sender will
not harm transport, and could potentially assist it, unless it is of
sufficient volume to result in the same type of impact as the "rogue
traffic generation" described above. However, it is possible that
replay of sufficiently old messages may cause receivers to think they
are "out of sync" with the sender and reset state, compromising the
transfer. Also, if sender transport data identifiers are reused
(object identifiers, FEC payload identifiers, etc), it is possible
Adamson, et al. Expires April 22, 2010 [Page 12]
Internet-Draft Security and RMT Protocols October 2009
that replay of old messages could corrupt data of a current transfer.
3.2.4.2. Replay of Receiver Messages
Replay of receiver messages are problematic for the NORM protocol,
because replay of NACK messages could cause the sender to
unnecessarily transmit repair information for an FEC coding block.
Similarly, the sender transmission rate might be manipulated by
replay of congestion control feedback messages from receivers in the
group. And the way that NORM senders estimate group round-trip
timing (GRTT) could allow a replay attack to manipulate the senders'
GRTT estimate to an unnecessarily large value, adding latency to the
reliable transport process.
4. General Security Goals
The term "security" is extremely vast and encompasses many different
meanings. The goal of this section is to clarify what "security"
means when considering the CDP defined in the IETF RMT working group.
However, the scope can also encompass additional applications, like
streaming applications. This section only focuses on the desired
general goals. The following sections will then discuss the possible
elementary services that will be required to fulfill these general
goals, as well as the underlying technological building blocks.
The possible final goals include, in decreasing order of importance:
o network protection: the goal is to protect the network from
attacks, no matter whether these attacks are voluntary (i.e.,
launched by one or several attackers) or non voluntary (i.e.,
caused by a misbehaving system, where "system" can designate a
building block, a protocol, an application, or a user);
o protocol protection: the goal is to protect the RMT protocol
itself, e.g., to avoid that a misbehaving receiver prevents other
receivers to get the content, no matter whether this is done
intentionally or not;
o content protection: to goal is to protect the content itself, for
instance to guaranty the integrity of the content, or to make sure
that only authorized clients can access the content;
o and user protection: the goal is often to protect the user
privacy.
Adamson, et al. Expires April 22, 2010 [Page 13]
Internet-Draft Security and RMT Protocols October 2009
4.1. Network Protection
Protecting the network is of course of primary importance. An
attacker should not be able to damage the whole infrastructure by
exploiting some features of the RMT protocol. Unfortunately, recent
past has shown that the multicast routing infrastructure is
relatively fragile, as well as the applications built on top of it.
Since the RMT protocols may use congestion control mechanisms to
regulate sender transmission rate, the protocol security features
should ensure that the sender may not be manipulated to transmit at
incorrect rates (most importantly not at an excessive rate) to any
parts of the receiver group. In the case of NORM, the security
mechanisms should ensure that the feedback suppression mechanisms are
protected to prevent badly-behaving network nodes from purposefully
causing feedback implosion. In the case of ALC, where layered
congestion control may be used via dynamic grou/layer membership,
this extends to considerations of excessive manipulation of the
multicast router control plane.
4.2. Protocol Protection
Protecting the protocols is also of importance, since the higher the
number of clients, the more serious the consequences of an attack.
This is all the more true as scalability is often one of the desired
goals of CDP. Ideally, receivers should be sufficiently isolated
from one another, so that a single misbehaving receiver does not
affect others. Similarly, an external attacker should not be able to
break the system, i.e., resulting in unreliable operation or delivery
of incorrect content.
4.3. Content Protection
The content itself should be protected when meaningful. This level
of security is often the concern of the content provider (and its
responsibility). For instance, in case of confidential (or non-free)
content, the typical solution consists in encrypting the content. It
can be done within the upper application, i.e., above the RMT
protocol, or within the transport system.
But other requirements may exist, like verifying the integrity of a
received object, or authenticating the sender of the received
packets. To that goal, one can rely on the use of building blocks
integrated within, or above, or beneath the RMT protocol.
One may also consider that offering the packet sender authentication
and content integrity services are basic requirements that should
fulfill any RMT system that operates within an open network, where
any attacker can easily inject spurious traffic in an ongoing NORM or
Adamson, et al. Expires April 22, 2010 [Page 14]
Internet-Draft Security and RMT Protocols October 2009
ALC session. In that case this goal is not the responsibility of the
content provider but the responsibility of the administrator who
deploys the RMT system itself.
4.4. Privacy
Finally the user should be protected, and more specifically its
privacy. In general, there is no privacy issue for data sender: the
data sender's address is announced to all prospective receivers prior
to their joins. Moreover receivers need to specify the source
address(es) as well as the IP multicast address in SSM communication
upon their subscription. The situation is different if we consider
receivers since their address should not be disclosed publicly.
Data receivers use IGMP or MLD protocols to notify their upstream
routers to join or leave IP multicast session. The recent IGMPv3
[RFC3376] and MLDv2 [RFC3810] do not adopt the "report suppression
mechanism". Report suppression makes the receiver host withdraw its
own report when the host hears a report scheduled to be sent from
other host joining the same group. Eliminating the report
suppression mechanism does not contribute to minimizing the number of
responses, but enables the router to keep track of host membership
status on a link. Due to this specification, operators who maintain
upstream routers that attach multicast data receiver can recognize
data receivers' addresses by tracing IGMP/MLD report messages.
Although such traced data may be useful for capacity planning or
accounting from operator's perspective, the detail information
including receivers' IP addresses should be carefully treated.
As described in Section 3.1.2, unauthorized users may spoof IGMP/MLD
query messages and trace receivers' addresses on the same LAN.
Currently, IGMP/MLD protocols do not protect this attack. It is
desired for these protocols to ignore invalid query messages and
provide receiver's privacy by some means.
5. Elementary Security Techniques
The goals defined in Section 4 will be fulfilled by means of
underlying security techniques, provided by one or several
technological building blocks. This section only focuses on these
elementary security techniques. Some general techniques
traditionally available are:
Adamson, et al. Expires April 22, 2010 [Page 15]
Internet-Draft Security and RMT Protocols October 2009
+-----------------+-------------------------------------------------+
| Technique | Goal |
+-----------------+-------------------------------------------------+
| packet | Enable session participants to verify that a |
| integrity | packet has not been inappropriately modified in |
| | transit. |
| packet source | Enable a receiver to verify the source of a |
| authentication | packet. |
| packet group | Enable a receiver to verify that a packet |
| authentication | originated or was modified only within the |
| | group and has not been modified by nonmembers |
| | in transit; Additionally, if attribution of any |
| | modifications by the group is required, certain |
| | group authentication mechanisms may provide |
| | this capability. |
| packet | Enable any third party to verify the source of |
| non-repudiation | a packet such that the source cannot repudiate |
| | having sent the packet. |
| packet | Enable a receiver to detect that a packet is |
| anti-replay | the same as a previously-received packet |
| object | Enable a receiver to verify the integrity of a |
| integrity | whole object. Such object integrity |
| | verification should be possible for any |
| | singular object or any composition of |
| | sub-objects which together constitute a larger |
| | object structure. |
| object source | Enable a receiver to verify the source of an |
| authentication | object. |
| object | Enable a source to guarantee that only |
| confidentiality | authorized receivers can access the object |
| | data. |
+-----------------+-------------------------------------------------+
General Security Techniques
Some additional techniques are specific to the RMT protocols:
+---------------+---------------------------------------------------+
| Technique | Goal |
+---------------+---------------------------------------------------+
| congestion | Prevent an attacker from modifying the congestion |
| control | control protocol normal behavior (e.g., by |
| security | reducing the transmission (NORM) or reception |
| | (ALC) rate, or on the opposite increasing this |
| | rate up to a point where congestion occurs) |
| group | Ensure that only authorized receivers (as defined |
| management | by a certain group management policy) join the |
| | RMT session and possibly inform the source |
Adamson, et al. Expires April 22, 2010 [Page 16]
Internet-Draft Security and RMT Protocols October 2009
| backward | Prevent a new group member to access the |
| group secrecy | information in clear sent to the group before he |
| | joined the group |
| forward group | Prevent a former group member to access the |
| secrecy | information in clear sent to the group after he |
| | left the group |
+---------------+---------------------------------------------------+
RMT-Specific Security Techniques
These techniques are usually achieved by means of one or several
technological building blocks. The target use case where the RMT
system will be deployed will greatly impact the choice of the
technological building block(s) used to provide these services, as
explained in Section 2.3.
6. Technological Building Blocks
Here is a list of techniques and building blocks that are likely to
fulfill one or several of the goals listed above:
o IPsec;
o Group MAC;
o Digital signatures;
o TESLA;
o SSM communication model;
Each of them is now quickly discussed. In particular we identify
what service it can offer, its limitations, and its field of
application (adequacy with respect to the CDP and the target use
case).
6.1. IPsec
6.1.1. Benefits
One direct approach using existing standards is to apply IPsec
[RFC2401] to achieve the following properties:
o source authentication and packet integrity (IPsec AH or ESP)
o confidentiality by means of encryption (IPsec ESP)
Adamson, et al. Expires April 22, 2010 [Page 17]
Internet-Draft Security and RMT Protocols October 2009
6.1.2. Requirements
It is expected that the approach to apply IPsec for reliable
multicast transport sessions is similar to that described for OSPFv3
security[RFC4552]. The following list proposes the IPsec
capabilities needed to support a similar approach to RMT protocol
security:
o Mode - Transport mode IPsec security is required;
o Selectors - source and destination addresses and ports, protocol.
o For some uses, preplaced, manual key support may be required to
support application deployment and operation. For automated key
management for group communication the Group Secure Association
Key Management Protocol (GSAKMP) described in [RFC4535] may be
used to emplace the keys for IPsec operation.
Note that a periodic rekeying procedure similar to that described in
RFC 4552 can also be applied with the additional benefit that the
reliable transport aspects of the CDP provide robustness to any
message loss that might occur due to ANY timing discrepancies among
the participants in the reliable multicast session.
6.1.3. Limitations
It should be noted that current IPsec implementations may not provide
the capability for anti-replay protection for multicast operation.
In the case of the NORM protocol, a sequence number is provided for
packet loss measurement to support congestion control operation.
This sequence number can also be used within a NORM implementation
for detecting duplicate (replayed) messages from sources (senders or
receivers) within the transport session group. In this way,
protection against replay attack can be achieved in conjunction with
the authentication and possibly confidentiality properties provided
by an IPsec encapsulation of NORM messages. NORM receivers generate
a very low volume of feedback traffic and it is expected that the 16-
bit sequence space provided by NORM will be sufficient for replay
attack protection. When a NORM session is long-lived, the limits of
the sender repair window are expected to provide protection from
replayed NACKs as they would typically be outside of the sender's
current repair window. It is suggested that IPsec implementations
that can provide anti-replay protection for IP Multicast traffic,
even when there are multiple senders within a group, be adopted. The
GSAKMP document has some discussion in this area.
Adamson, et al. Expires April 22, 2010 [Page 18]
Internet-Draft Security and RMT Protocols October 2009
6.2. Group MAC
6.2.1. Benefits
The use of Group MAC (Message Authentication Codes) within the CDP
[SimpleAuth] is a simple solution to provide a loss tolerant group
authentication/integrity service for all the packets exchanged within
a session (i.e., the packets generated by the session's sender and
the session's receivers). This scheme is easy to deploy since it
only requires that all the group members share a common secret key.
Because Group MAC heavily relies on fast symmetric cryptographic
building blocks, CPU processing remains limited both at the sender
and receiver sides, which makes it suitable for high data rate
transmissions, and/or lightweight terminals. Finally, the
transmission overhead remains limited.
6.2.2. Requirements
This scheme only requires that all the group members share a common
secret key, possibly associated to a re-keying mechanism (e.g., each
time the group membership changes, or on a periodic basis).
6.2.3. Limitations
This scheme cannot protect against attacks coming from inside the
group, where a group member impersonates the sender and sends forged
messages to other receivers. It only provides a group-level
authentication/integrity service, unlike the TESLA and Digital
Signature schemes. Note that the Group MAC and Digital Signature
schemes can be advantageously used together, as explained in
[SimpleAuth].
6.3. Digital Signatures
6.3.1. Benefits
The use of Digital Signatures within the CDP [SimpleAuth] is a simple
solution to provide a loss-tolerant authentication/integrity service
for all the packets exchanged within a session (i.e., the packets
generated by the session's sender and the session's receivers). This
scheme is easy to deploy since it only requires that the participants
know the packet sender's public key, which can be achieved with
either Public Key Infrastructure (PKI) or by preplacement of these
keys.
Adamson, et al. Expires April 22, 2010 [Page 19]
Internet-Draft Security and RMT Protocols October 2009
6.3.2. Requirements
This scheme is easy to deploy since it requires only that the
participants know the packet sender's public key, which can be
achieved with either PKI or by preplacement of these keys.
6.3.3. Limitations
When RSA [RsaPaper] asymmetric cryptography is used, the digital
signatures approach has two major shortcomings:
o it is limited by high computational costs, especially at the
sender, and
o it is limited by high transmission overheads.
This scheme is well suited to low data rate flows, when transmission
overheads are not a major issue. For instance it can be used as a
complement to TESLA for the feedback traffic coming from the
session's receivers. The use of ECC ("Elliptic Curve Cryptography")
significantly relaxes these constraints, especially when seeking for
higher security levels. For instance, the following key size provide
equivalent security:
+--------------------+--------------+--------------+
| Symmetric Key Size | RSA Key Size | ECC Key Size |
+--------------------+--------------+--------------+
| 80 bits | 1024 bits | 160 bits |
| 112 bits | 2048 bits | 224 bits |
+--------------------+--------------+--------------+
However in some cases, the Intellectual Property Rights (IPR)
considerations for ECC may limit its use, so the other techniques are
presented here as well. Note that the Group MAC and Digital
Signature schemes can be advantageously used together, as explained
in [SimpleAuth].
6.4. TESLA
6.4.1. Benefits
The use of TESLA [I-D.ietf-msec-tesla-for-alc-norm] within the CDP
offers a loss tolerant, lightweight, authentication/integrity service
for the packets generated by the session's sender. Depending on the
time synchronization and bootstrap methods used, TESLA can be
compatible with massively scalable sessions. Because TESLA heavily
relies on fast symmetric cryptographic building blocks, CPU
processing remains limited both at the sender and receiver sides,
Adamson, et al. Expires April 22, 2010 [Page 20]
Internet-Draft Security and RMT Protocols October 2009
which makes it suitable for high data rate transmissions, and/or
lightweight terminals. Finally, the transmission overhead remains
limited.
6.4.2. Requirements
The security offered by TESLA relies heavily on time. Therefore the
session's sender and each receiver need to be loosely synchronized in
a secure way. To that purpose, several methods exist, depending on
the use case: direct time synchronization (which requires a
bidirectional transport channel), using a secure Network Time
Protocol (NTP) [RFC1305] infrastructure (which also requires a
bidirectional transport channel), or a Global Positioning System
(GPS) device, or a clock with a time-drift that is negligible in
front of the TESLA time accuracy requirements.
The various bootstrap parameters must also be communicated to the
receivers, using either an in-band or out-of-band mechanism,
sometimes requiring bidirectional communications. So, depending on
the time synchronization scheme and the bootstrap mechanism method,
TESLA can be used with either bidirectional or unidirectional
transport channels.
6.4.3. Limitations
One limitation is that TESLA does not protect the packets that are
generated by receivers, for instance the feedback packets of NORM.
These packets must be protected by other means.
Another limitation is that TESLA requires some buffering capabilities
at the receivers in order to enable the delayed authentication
feature. This is not considered though as a major issue in the
general case (e.g., FEC decoding of objects within an ALC session
already requires some buffering capabilities, that often exceed that
of TESLA), but it might be one in case of embedded environments.
6.5. Source-Specific Multicast
Source-Specific Multicast (SSM) [RFC3569], [RFC4607] amends the
classical Any-Source Multicast (ASM) model by creating logical IP
multicast "channels" that are defined by the multicast destination
address _and_ the specific source address(es). Thus for a given
"channel", only the specific source(s) can inject packets that are
distributed to the receivers. This form of multicast has group
management benefits since a source can independently control the
"channels" it creates.
Adamson, et al. Expires April 22, 2010 [Page 21]
Internet-Draft Security and RMT Protocols October 2009
6.5.1. Requirements
Use of SSM requires that the network intermediate systems explicitly
support it. Additionally, hosts operating systems are required to
support the IGMPv3/MLDv2 extensions for SSM, and the CDP
implementations need to support the IGMPv3/MLDv2 API, including
management of the <srcAddr; dstMcastAddr> "channel" identifiers.
6.5.2. Limitations
CDP such as NORM that use signaling from receivers to multicast
senders will need to use unicast addressing for feedback messages.
In the case of NORM, its timer-based feedback suppression requires
support of the sender NORM_CMD(REPAIR_ADV) message to control
receiver feedback. In some topologies, use of unicast feedback may
require some additional latency (increased backoff factor) for safe
operation. The security of the unicast feedback from the receivers
to sender will need to be addressed separately since the IP multicast
model, including SSM, does not provide the sender knowledge of
authorized group members.
6.5.3. Source-Based and Receiver-Based Attacks
The security threats are categorized into "source-based" and
"receiver-based" attacks [RFC4609]. In short, the former is a DoS
attack against the multicast networks, in which data is sent to
numerous and random group addresses, and the latter is a DoS attack
against multicast routers, in which innumerable IGMP/MLD joins are
sent from a client.
Regarding source-based attack, there are some security benefits in
SSM. Since data-plane traffic for an SSM "channel" is limited to
that of a single, specific source address, it is possible that
network intermediate systems may impose mechanism that prevent
injection of traffic to the group from inappropriate (perhaps
malicious) nodes. This can reduce the risk for denial-of-service and
some of the other attacks described in this document. While SSM
alone is not a complete security solution, it can simplify secure RMT
operation.
On the contrary, SSM is not robust against receiver-based attack. An
SSM capable router constructs a Shortest-Path Tree (SPT) with no
shared tree coordination. Therefore, even if a host triggers invalid
or unavailable channel subscriptions, the upstream router starts
establishing all SPTs with no intellectual decision. What is worse
is that these multicast routers cannot recognize the original router
that is attacked and cannot stop the attack itself.
Adamson, et al. Expires April 22, 2010 [Page 22]
Internet-Draft Security and RMT Protocols October 2009
6.6. Summary
The following table summarizes the pros/cons of each authentication/
integrity scheme used at application/transport level (where "-" means
bad, "0" means neutral, and "+" means good):
+------------------+-------------+-------------+------------+-------+
| | RSA Digital | ECC Digital | Group MAC | TESLA |
| | Signature | Signature | | |
+------------------+-------------+-------------+------------+-------+
| True | Yes | Yes | No (group | Yes |
| authentication | | | security) | |
| and integrity | | | | |
| Immediate | Yes | Yes | Yes | No |
| authentication | | | | |
| Processing load | - | 0 | + | + |
| Transmission | - | 0 | + | + |
| overhead | | | | |
| Complexity | + | + | + | - |
+------------------+-------------+-------------+------------+-------+
7. Security Infrastructure
Deploying the elementary technological building blocks often requires
that a security infrastructure exists. Such security infrastructure
can provide:
o Public Key Infrastructure (PKI) for trusted third party vetting
of, and vouching for, user identities. PKI also allows the
binding of public keys to users, usually by means of certificates.
o Group Key Management with rekeying schemes that are either
periodic or triggered by some higher level event. It is required
in particular when the group is dynamic and forward/backward
secrecy are important. This is also required to improve the
scalability of the CDP (since key management is done
automatically, using a key server topology), or the security
provided by the CDP (since the underlying cryptographic keys will
be changed frequently)
It is expected that some CDP deployments may use existing client-
server security infrastructure models so that receivers may acquire
any necessary security material and be authenticated or validated as
needed for group participation. Then, the reliable delivery of
session data content will be provided via the applicable RMT
protocols. Note that in this case the security infrastructure itself
may limit the scalability of the group size or other aspects of
Adamson, et al. Expires April 22, 2010 [Page 23]
Internet-Draft Security and RMT Protocols October 2009
reliable multicast transfer. The IETF Multicast Security (MSEC)
Working Group has developed some protocols that can be applied to
achieve more scalable and effective group communication security
infrastructure[RFC4046]. It is encouraged that these mechanisms be
considered in the development of security for CDP.
8. New Threats Introduced by the Security Scheme Itself
Introducing a security scheme, as a side effect, can sometimes
introduce new security threats. For instance, signing all packets
with asymmetric cryptographic schemes (to provide a source
authentication/content integrity/anti-replay service) opens the door
to DoS attacks. Indeed, verifying asymmetric-based cryptographic
signatures is a CPU intensive task. Therefore an attacker can easily
overload a receiver (or a sender in case of NORM) by injecting a
significant number of faked packets.
9. Consequences for the RMT and MSEC Working Group
To meet the goals outlined in this document, it is expected that the
RMT and MSEC Working Groups may need to develop some supporting
protocol security mechanisms. It is also possible to cooperate with
the Multicast Backbone (MBONE) Deployment (MBONED) Working Group for
defining operational considerations.
9.1. RMT Transport Message Security Encapsulation Header
An alternative approach to using IPsec to provide the necessary
properties to protect RMT protocol operation from the application
attacks described earlier, is to extend the RMT protocol message set
to include a message encapsulation option. This encapsulation header
could be used to provide authentication, confidentiality, and anti-
replay protection as needed. Since this would be independent of the
IP layer, the header might need to provide a source identifier to be
used as a "selector" for recalling security state (including
authentication certificate(s), sequence state, etc) for a given
message. In the case of the NORM protocol, a "NormNodeId" field
exists that could be used for this purpose. In the case of ALC, the
security encapsulation mechanism would need to add this function.
The security encapsulation mechanism, although resident "above" the
IP layer, could use GSAKMP [RFC4535] or a similar approach for
automated key management.
Adamson, et al. Expires April 22, 2010 [Page 24]
Internet-Draft Security and RMT Protocols October 2009
10. Security Considerations
This document is a general discussion of security for the RMT
protocol family. But specific security considerations are not
applicable as this document does not introduce any new techniques.
11. Acknowledgments
The authors would like acknowledge Magnus Westerlund for stimulating
the working group activity in this area. Additionally George Gross
and Ran Atkinson contributed many ideas to the discussion here.
12. References
12.1. Normative References
[I-D.ietf-msec-tesla-for-alc-norm]
Roca, V., Francillon, A., and S. Faurite, "Use of TESLA in
the ALC and NORM Protocols",
draft-ietf-msec-tesla-for-alc-norm-07 (work in progress),
December 2008.
[I-D.ietf-rmt-flute-revised]
Luby, M., Lehtonen, R., Roca, V., and T. Paila, "FLUTE -
File Delivery over Unidirectional Transport",
draft-ietf-rmt-flute-revised-06 (work in progress),
September 2008.
[I-D.ietf-rmt-pi-alc-revised]
Luby, M., Watson, M., and L. Vicisano, "Asynchronous
Layered Coding (ALC) Protocol Instantiation",
draft-ietf-rmt-pi-alc-revised-06 (work in progress),
November 2008.
[I-D.ietf-rmt-pi-norm-revised]
Adamson, B., Bormann, C., London, U., and J. Macker,
"NACK-Oriented Reliable Multicast Protocol",
draft-ietf-rmt-pi-norm-revised-08 (work in progress),
December 2008.
[RFC1112] Deering, S., "Host extensions for IP multicasting", STD 5,
RFC 1112, August 1989.
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.
Adamson, et al. Expires April 22, 2010 [Page 25]
Internet-Draft Security and RMT Protocols October 2009
[RFC2401] Kent, S. and R. Atkinson, "Security Architecture for the
Internet Protocol", RFC 2401, November 1998.
[RFC3376] Cain, B., Deering, S., Kouvelas, I., Fenner, B., and A.
Thyagarajan, "Internet Group Management Protocol, Version
3", RFC 3376, October 2002.
[RFC3569] Bhattacharyya, S., "An Overview of Source-Specific
Multicast (SSM)", RFC 3569, July 2003.
[RFC3810] Vida, R. and L. Costa, "Multicast Listener Discovery
Version 2 (MLDv2) for IPv6", RFC 3810, June 2004.
[RFC4046] Baugher, M., Canetti, R., Dondeti, L., and F. Lindholm,
"Multicast Security (MSEC) Group Key Management
Architecture", RFC 4046, April 2005.
[RFC4535] Harney, H., Meth, U., Colegrove, A., and G. Gross,
"GSAKMP: Group Secure Association Key Management
Protocol", RFC 4535, June 2006.
[RFC4552] Gupta, M. and N. Melam, "Authentication/Confidentiality
for OSPFv3", RFC 4552, June 2006.
[RFC4607] Holbrook, H. and B. Cain, "Source-Specific Multicast for
IP", RFC 4607, August 2006.
[RFC4609] Savola, P., Lehtonen, R., and D. Meyer, "Protocol
Independent Multicast - Sparse Mode (PIM-SM) Multicast
Routing Security Issues and Enhancements", RFC 4609,
October 2006.
[RFC4654] Widmer, J. and M. Handley, "TCP-Friendly Multicast
Congestion Control (TFMCC): Protocol Specification",
RFC 4654, August 2006.
[RFC5052] Watson, M., Luby, M., and L. Vicisano, "Forward Error
Correction (FEC) Building Block", RFC 5052, August 2007.
[SimpleAuth]
Roca, V., "Simple Authentication Schemes for the ALC and
NORM Protocols", Internet
Draft draft-ietf-rmt-simple_auth-for-alc-norm-00.txt (work
in progress), October 2008.
Adamson, et al. Expires April 22, 2010 [Page 26]
Internet-Draft Security and RMT Protocols October 2009
12.2. Informative References
[Neumann05]
Neumann, C., Roca, V., and R. Walsh, "Large Scale Content
Distribution Protocols", ACM Computer Communications
Review (CCR) Vol. 35 No. 5, October 2005.
[RFC1305] Mills, D., "Network Time Protocol (Version 3)
Specification, Implementation", RFC 1305, March 1992.
[RsaPaper]
Rivest, R., Shamir, A., and L. Adleman, "A Method for
Obtaining Digital Signatures and Public-Key
Cryptosystems", Communications of the ACM 21, pp. 120-126,
1978.
Authors' Addresses
Brian Adamson
Naval Research Laboratory
Washington, DC 20375
USA
Email: adamson@itd.nrl.navy.mil
URI: http://cs.itd.nrl.navy.mil
Vincent Roca
INRIA
Montbonnot 38334
France
Email: vincent.roca@inria.fr
URI: http://planete.inrialpes.fr/~roca/
Hitoshi Asaeda
Keio University
5322 Endo
Fujisawa, Kanagawa 252-8520
Japan
Email: asaeda@wide.ad.jp
URI: http://www.sfc.wide.ad.jp/~asaeda/
Adamson, et al. Expires April 22, 2010 [Page 27]
| PAFTECH AB 2003-2026 | 2026-04-22 23:25:54 |