Network Working Group Stanislav Shalunov
Internet Draft Internet2
Expiration Date: December 2002 Benjamin Teitelbaum
Advanced Network & Services and Internet2
June 2002
A One-way Active Measurement Protocol Requirements
<draft-ietf-ippm-owdp-reqs-02.txt>
1. Status of this Memo
This document is an Internet-Draft and is in full conformance with
all provisions of Section 10 of RFC2026.
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 memo provides information for the Internet community. This memo
does not specify an Internet standard of any kind. Distribution of
this memo is unlimited.
2. Abstract
With growing availability of good time sources to network nodes, it
becomes increasingly possible to measure one-way IP performance
metrics with high precision. To do so in an interoperable manner, a
common protocol for such measurements is required. This document
specifies requirements for a one-way active measurement protocol
(OWAMP) standard. The protocol can measure one-way delay, as well as
other unidirectional characteristics, such as one-way loss.
Shalunov and Teitelbaum [Page 1]
INTERNET-DRAFT One-way Delay Protocol Requirements June 2002
3. Motivations and Goals
The IETF IP Performance Metrics (IPPM) working group has proposed
draft standard metrics for one-way packet delay [RFC2679] and loss
[RFC 2680] across Internet paths. Although there are now several
measurement platforms that implement the collection of these metrics
([CQOS], [BRIX], [RIPE], [SURVEYOR]), there is not currently a
standard for interoperability. This requirements document is aimed
at defining a protocol that allows users to do measurements using
devices from different vendors at both ends and get meaningful
results.
With the increasingly wide availability of affordable global
positioning system (GPS) and CDMA based time sources, hosts
increasingly have available to them time sources that allow hosts to
time-stamp packets with accuracies substantially better than the
delays seen on the Internet--either directly or through their
proximity to NTP primary (stratum 1) time servers. By standardizing
a technique for collecting IPPM one-way active measurements, we hope
to create an environment where these metrics may be collected across
a far broader mesh of Internet paths than is currently possible. One
particularly compelling vision is of widespread deployment of open
one-way active measurement beacons that would make measurements of
one-way delay as commonplace as measurements of round-trip time are
today using ICMP-based tools like ping. Even without very accurate
timestamps one can measure characteristics such as loss with quality
acceptable for many practical purposes, e.g., network operations.
To support interoperability between alternative OWAMP implementations
and make possible a world where "one-way ping" could become
commonplace, a standard is required that specifies how test streams
are initiated, how test packets are exchanged, and how test results
are retrieved. Detailed functional requirements are given in the
subsequent section.
4. Functional Requirements
The protocol(s) should provide the ability to measure, record, and
distribute the results of measurements of one-way singleton network
characteristics such as characteristics defined in [RFC2679] and
[RFC2680]. Result reporting, sampling, and time stamps are to be
within the framework of [RFC2330].
It should be possible to measure arbitrary one-way singleton
characteristics (e.g., loss, median delay, mean delay, jitter, 90th
percentile of delay, etc.); this is achieved by keeping all the raw
data for post-processing by the final data consumer, as specified in
Shalunov and Teitelbaum [Page 2]
INTERNET-DRAFT One-way Delay Protocol Requirements June 2002
section 4.1. Since RFC2679 and RFC2680 standardize metrics based on
Poisson sampling processes, Poisson streams must be supported by the
protocol(s).
Non-singleton characteristics (such as those related to trains of
packets, back-to-back tuples, and so forth) and application traffic
simulation need not be addressed. However, they may be addressed if
considered practical and not in contradiction to other design goals.
4.1. Keeping All Data for Post-processing
To facilitate the broadest possible use of obtained measurement
results, the protocol(s) should not necessitate any required post-
processing. (This does not apply to implementation details such as
converting timestamps from ticks since midnight into a canonical form
or applying calibration constants; such details should naturally be
hidden.) All data obtained during a measurement session should be
available after the session is finished if desired by the data
consumer so that various characteristics can be computed from the raw
data using arbitrary algorithms.
4.2. Result Distribution
A means to distribute measurement results (between hosts
participating in a measurement session and beyond) should be
provided. Since there can exist a wide variety of scenarios as to
where the final data destination should be, these should be invoked
separately from measurement requests (e.g., receiver should not have
to automatically send measurement results to the sender, since it may
be the receiver or a third host that are the ultimate data
destination).
At the same time, ability to transfer results directly to their
destination (without need for potentially large intermediate
transfers) should be provided.
4.3. Protocol Separation
Since measurement session setup and the actual measurement session
(i) are different tasks; (ii) require different levels of
functionality, flexibility, and implementation effort; (iii) may need
to run over different transport protocols, there should exist two
protocols: one for conducting the actual measurement session and
another for session setup/teardown/confirmation/retrieval. These
protocols are further referred to as OWAMP-Test and OWAMP-Control,
Shalunov and Teitelbaum [Page 3]
INTERNET-DRAFT One-way Delay Protocol Requirements June 2002
respectively.
It should be possible to use devices that only support OWAMP-Test but
not OWAMP-Control to conduct measurement sessions (such devices will
necessarily need to support one form of session setup protocol or the
other, but it doesn't have to be known to external parties).
OWAMP-Control would thus become a common protocol for different
administrative domains, which may or may not use it for session setup
internally.
4.4. Test Protocol
The test protocol needs to be implemented on all measurement nodes
and should therefore have the following characteristics:
+ Be lightweight and easy to implement.
+ Be suitable for implementation on a wide range of measurement
nodes.
+ Allow UDP as the transport protocol, since the protocol needs to
be able to measure individual packet delivery times and has to run
on various machines (see the section "Support for Measurements
with Different Packet Types" below for further discussion).
+ Support varying packet sizes and network services (e.g., DSCP
marking), as negotiated by OWAMP-Control.
+ Be as simple as possible, but no simpler than necessary to
implement requirements set forth in this document; the OWAMP-Test
packet format should include only universally meaningful fields,
and minimum number of them.
+ If practical, it should be possible to generate OWAMP-Test packets
small enough, so that when encapsulated, each fits inside a single
ATM cell.
+ Data needed to calculate experimental errors on the final result
should be included in every OWAMP-Test packet.
Shalunov and Teitelbaum [Page 4]
INTERNET-DRAFT One-way Delay Protocol Requirements June 2002
4.5. Control Protocol
Control protocol needs to provide the capability to:
+ authenticate peers to each other using a common authentication
method that doesn't require building any new authentication
infrastructure, such as user ID and a shared secret;
+ schedule zero or more OWAMP-Test sessions (which do not have to be
between the peers of OWAMP-Control conversation);
+ start sessions simultaneously or at a pre-scheduled per-session
times;
+ retrieve OWAMP-Test session results (of OWAMP-Test sessions
scheduled in the current and other OWAMP-Control sessions);
+ confirm graceful completion of sessions and allow either side to
abort a session prematurely.
The OWAMP-Control design should not preclude the ability to record
extended periods of losses. It should always provide peers with the
ability to distinguish between network and peer failures.
4.6. Support for Measurements with Different Packet Types
Since the notion of a packet of type P from [RFC2330], section 13
doesn't always imply precise definition of packet type, some
decisions narrowing the scope of possible packet types need to be
made at measurement protocol design stage. Further, measurement with
packets of certain types, while feasible in more closed settings than
those implied by OWAMP, become very hard to perform in an open inter-
domain fashion (e.g., measurements with particular packets with
broken IP checksum or particular loose source routing options).
In addition, very general packet type specification could result in
several problems:
+ Many OWAMP-Test speakers will be general purpose computers with a
multitasking operating system that includes a socket interface.
These will inevitably have higher losses when listening to raw
network traffic. Raw sockets will induce higher loss rate than
one would have with UDP measurements.
Shalunov and Teitelbaum [Page 5]
INTERNET-DRAFT One-way Delay Protocol Requirements June 2002
+ It's not at all clear (short of standardizing tcpdump syntax) how
to describe formally the filter that a receiver should use to
listen for test traffic.
+ Suppose an identity of an authenticated user becomes compromised.
Now the attacker could use that to run TCP sessions to the rlogin
port of machines around servers that trust this user to perform
measurements (or, less drastically, to send spam from that
network). The ability to perform measurements is transformed into
an ability to generate arbitrary traffic on behalf of all the
senders an OWAMP-Control server controls.
+ Carefully crafted packets could cause disruption to some link-
layer protocols. Implementors can't know what to disallow
(scrambling is different for different link-layer technologies).
It appears that allowing one to ask a measurement server to generate
arbitrary packets becomes an unmanageable security hole and a
formidable specification and implementation hurdle.
For these reasons, we only require OWAMP to support a small subspace
of the whole packet type space. Namely, it should be possible to
conduct measurements with a given Differentiated Services Codepoint
(DSCP) [RFC2474] or a given Per Hop Behavior Identification Code (PHB
ID) [RFC2836].
5. Scalability
While some measurement architecture designs have inherent scalability
problems (e.g., a full mesh of always-on measurements among N
measurement nodes requires O(N^2) total resources, such as storage
space and link capacity), OWAMP itself should not exaggerate the
problem or make it impossible (where it is in principle possible) to
design other architectures that are free of scalability deficiencies.
It is the protocol user's responsibility to decide how to use the
protocol and which measurements to conduct.
Shalunov and Teitelbaum [Page 6]
INTERNET-DRAFT One-way Delay Protocol Requirements June 2002
6. Security Considerations
6.1. Modes of Operation
Since the protocol(s) will be used in widely varying circumstances
using widely varying equipment, it is necessary be able to support
varying degrees of security modes of operation. The parameters to be
considered include: confidentiality, data origin authentication,
integrity and replay protection.
It should be possible to operate in a completely "open" mode, where
no cryptographic security mechanisms are used. We call this
"unauthenticated mode". In this mode, care must be taken to avoid
denial-of-service attacks.
It should also be possible to operate in a mode where all security
mechanisms are enabled and security objectives are realized to the
fullest extent possible. We call this "encrypted mode".
Since timestamp encryption takes a certain amount of time, which may
be hard to predict on some devices (with a time-sharing OS), a mode
should be provided that is similar to encrypted mode, but in which
timestamps are not encrypted. In this mode, all security properties
of the encrypted mode that can be retained without timestamp
encryption should be present. We call this "authenticated mode".
To make implementation more manageable, the number of other options
and modes should be kept to the absolute practical minimum. Where
choosing a single mechanism for achieving anything related to
security is possible, such choice should be made at specification
phase and be put into the standard.
6.2. Being Hard to Interfere with by Applying Special Treatment to
Measurement Packets
The design of the protocol should make it possible to run sessions
that would make it very difficult for any intermediate party to make
results appear better than they would be if no interference was
attempted.
This is different from cryptographic assurance of data integrity,
because one can manipulate the results without changing any data in
the packets. For example, if OWAMP-Test packets are easy to identify
(e.g., they all come to a well-known port number), an intermediate
party might place OWAMP-Test traffic into a priority queue at a
congested link thus ensuring that the results of the measurement
Shalunov and Teitelbaum [Page 7]
INTERNET-DRAFT One-way Delay Protocol Requirements June 2002
appear better than what would be experienced by other traffic. It
should not be easy for intermediate parties to identify OWAMP-Test
packets (just as it should not be easy for restaurants to identify
restaurant critics).
6.3. Secrecy/Confidentiality
It should be possible to make it infeasible for any outside party
without knowledge of the shared secret being used to learn what
information is exchanged using OWAMP-Control by inspecting an OWAMP-
Control stream or actively modifying it.
(It is recognized that some information will inevitably leak from the
mere fact of communication and from the presence and timing of
concurrent and subsequent OWAMP-Test traffic.)
6.4. Authentication
It should be possible to authenticate peers to each other using a
user ID and a shared secret. It should be infeasible for any
external party without knowledge of the shared secret to obtain any
information about it by observing, initiating, or modifying protocol
transactions.
It should also be infeasible for such party to use any information
obtained by observing, modifying or initiating protocol transactions
to impersonate (other) valid users.
6.5. Integrity
So that it is possible to detect any interference during a
conversation (other than the detention of some messages), facility
must be provided to authenticate each message of the control
protocol, its attribution to a given session, and its exact placement
in the sequence of control protocol exchanges.
It must also be possible to authenticate each message of the test
protocol and its attribution to a specific session, so that
modifications of OWAMP-Test messages can be detected. It must be
possible to do this in a fashion that does not require timestamps
themselves to be encrypted; in this case, security properties are
valid only when an attacker cannot observe valid traffic between the
OWAMP-Test sender and receiver.
Shalunov and Teitelbaum [Page 8]
INTERNET-DRAFT One-way Delay Protocol Requirements June 2002
7. IANA Considerations
Relevant IANA considerations will be placed into the protocol
specification document itself, and not into the requirements
document.
8. References
[BRIX] Brix 1000 Verifier,
http://www.brixnet.com/products/brix1000.html
[CQOS] CQOS Home Page, http://www.cqos.com/
[RFC2330] V. Paxson, G. Almes, J. Mahdavi, M. Mathis, "Framework for
IP Performance Metrics", RFC 2330, May 1998.
[RFC2474] K. Nichols, S. Blake, F. Baker, D. Black, "Definition of
the Differentiated Services Field (DS Field) in the IPv4 and
IPv6 Headers", RFC 2474, December 1998.
[RFC2679] G. Almes, S. Kalidindi, and M. Zekauskas, "A One-way Delay
Metric for IPPM", RFC 2679, September 1999.
[RFC2680] G. Almes, S. Kalidindi, and M. Zekauskas, "A One-way Packet
Loss Metric for IPPM", RFC 2680, September 1999.
[RFC2836] S. Brim, B. Carpenter, F. Le Faucheur, "Per Hop Behavior
Identification Codes", RFC 2836, May 2000.
[RIPE] RIPE NCC Test-Traffic Measurements home,
http://www.ripe.net/test-traffic/
[SURVEYOR] Surveyor Home Page, http://www.advanced.org/surveyor/
9. Authors' Addresses
Stanislav Shalunov
Internet2
200 Business Park Drive, Suite 307
Armonk, NY 10504
USA
Phone: +1 914 765 1182
EMail: shalunov@internet2.edu
Shalunov and Teitelbaum [Page 9]
INTERNET-DRAFT One-way Delay Protocol Requirements June 2002
Benjamin Teitelbaum
Advanced Network & Services
200 Business Park Drive, Suite 307
Armonk, NY 10504
USA
Phone: +1 914 765 1118
EMail: ben@advanced.org
Expiration date: December 2002
Shalunov and Teitelbaum [Page 10]
