Network Working Group K. Hedayat Request for Comments: 5357 Brix Networks Category: Standards Track R. Krzanowski Verizon A. Morton AT&T Labs K. Yum Juniper Networks J. Babiarz Nortel Networks October 2008 A Two-Way Active Measurement Protocol (TWAMP) Status of This Memo This document specifies an Internet standards track protocol for the Internet community, and requests discussion and suggestions for improvements. Please refer to the current edition of the "Internet Official Protocol Standards" (STD 1) for the standardization state and status of this protocol. Distribution of this memo is unlimited. Abstract The One-way Active Measurement Protocol (OWAMP), specified in RFC 4656, provides a common protocol for measuring one-way metrics between network devices. OWAMP can be used bi-directionally to measure one-way metrics in both directions between two network elements. However, it does not accommodate round-trip or two-way measurements. This memo specifies a Two-Way Active Measurement Protocol (TWAMP), based on the OWAMP, that adds two-way or round-trip measurement capabilities. The TWAMP measurement architecture is usually comprised of two hosts with specific roles, and this allows for some protocol simplifications, making it an attractive alternative in some circumstances. Hedayat, et al. Standards Track [Page 1] RFC 5357 Two-Way Active Measurement Protocol October 2008 Table of Contents 1. Introduction ....................................................2 1.1. Relationship of Test and Control Protocols .................3 1.2. Logical Model ..............................................3 1.3. Pronunciation Guide ........................................4 2. Protocol Overview ...............................................5 3. TWAMP-Control ...................................................6 3.1. Connection Setup ...........................................6 3.2. Integrity Protection .......................................7 3.3. Values of the Accept Field .................................7 3.4. TWAMP-Control Commands .....................................7 3.5. Creating Test Sessions .....................................8 3.6. Send Schedules ............................................10 3.7. Starting Test Sessions ....................................10 3.8. Stop-Sessions .............................................10 3.9. Fetch-Session .............................................12 4. TWAMP-Test .....................................................12 4.1. Sender Behavior ...........................................12 4.1.1. Packet Timings .....................................12 4.1.2. Packet Format and Content ..........................12 4.2. Reflector Behavior ........................................13 4.2.1. TWAMP-Test Packet Format and Content ...............14 5. Implementers' Guide ............................................20 6. Security Considerations ........................................20 7. Acknowledgements ...............................................21 8. IANA Considerations ............................................21 8.1. Registry Specification ....................................22 8.2. Registry Management .......................................22 8.3. Experimental Numbers ......................................22 8.4. Initial Registry Contents .................................22 9. Internationalization Considerations ............................22 Appendix I - TWAMP Light (Informative) ............................23 Normative References ..............................................24 Informative References ............................................24 1. Introduction The Internet Engineering Task Force (IETF) has completed a Proposed Standard for the round-trip delay [RFC2681] metric. The IETF has also completed a protocol for the control and collection of one-way measurements, the One-way Active Measurement Protocol (OWAMP) [RFC4656]. However, OWAMP does not accommodate round-trip or two-way measurements. Two-way measurements are common in IP networks, primarily because synchronization between local and remote clocks is unnecessary for round-trip delay, and measurement support at the remote end may be Hedayat, et al. Standards Track [Page 2] RFC 5357 Two-Way Active Measurement Protocol October 2008 limited to a simple echo function. However, the most common facility for round-trip measurements is the ICMP Echo Request/Reply (used by the ping tool), and issues with this method are documented in Section 2.6 of [RFC2681]. This memo specifies the Two-Way Active Measurement Protocol, or TWAMP. TWAMP uses the methodology and architecture of OWAMP [RFC4656] to define an open protocol for measurement of two-way or round-trip metrics (henceforth in this document the term two-way also signifies round-trip), in addition to the one-way metrics of OWAMP. TWAMP employs time stamps applied at the echo destination (reflector) to enable greater accuracy (processing delays can be accounted for). The TWAMP measurement architecture is usually comprised of only two hosts with specific roles, and this allows for some protocol simplifications, making it an attractive alternative to OWAMP in some circumstances. 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 RFC 2119 [RFC2119]. 1.1. Relationship of Test and Control Protocols Similar to OWAMP [RFC4656], TWAMP consists of two inter-related protocols: TWAMP-Control and TWAMP-Test. The relationship of these protocols is as defined in Section 1.1 of OWAMP [RFC4656]. TWAMP- Control is used to initiate, start, and stop test sessions, whereas TWAMP-Test is used to exchange test packets between two TWAMP entities. 1.2. Logical Model The role and definition of the logical entities are as defined in Section 1.2 of OWAMP [RFC4656] with the following exceptions: - The Session-Receiver is called the Session-Reflector in the TWAMP architecture. The Session-Reflector has the capability to create and send a measurement packet when it receives a measurement packet. Unlike the Session-Receiver, the Session-Reflector does not collect any packet information. - The Server is an end system that manages one or more TWAMP sessions, and is capable of configuring per-session state in the endpoints. However, a Server associated with a Session-Reflector would not have the capability to return the results of a test session, and this is a difference from OWAMP. Hedayat, et al. Standards Track [Page 3] RFC 5357 Two-Way Active Measurement Protocol October 2008 - The Fetch-Client entity does not exist in the TWAMP architecture, as the Session-Reflector does not collect any packet information to be fetched. Consequently, there is no need for the Fetch- Client. An example of possible relationship scenarios between these roles is presented below. In this example, different logical roles are played on different hosts. Unlabeled links in the figure are unspecified by this document and may be proprietary protocols. +----------------+ +-------------------+ | Session-Sender |<-TWAMP-Test-->| Session-Reflector | +----------------+ +-------------------+ ^ ^ | | | | | | | +----------------+<----------------+ | | Server | | +----------------+ | ^ | | | TWAMP-Control | | v v +----------------+ | Control-Client | +----------------+ As in OWAMP [RFC4656], different logical roles can be played by the same host. For example, in the figure above, there could actually be two hosts: one playing the roles of Control-Client and Session- Sender, and the other playing the roles of Server and Session- Reflector. This example is shown below. +-----------------+ +-------------------+ | Control-Client |<--TWAMP-Control-->| Server | | | | | | Session-Sender |<--TWAMP-Test----->| Session-Reflector | +-----------------+ +-------------------+ 1.3. Pronunciation Guide The acronym OWAMP is usually pronounced in two syllables, Oh-wamp. The acronym TWAMP is also pronounced in two syllables, Tee-wamp. Hedayat, et al. Standards Track [Page 4] RFC 5357 Two-Way Active Measurement Protocol October 2008 2. Protocol Overview The Two-Way Active Measurement Protocol is an open protocol for measurement of two-way metrics. It is based on OWAMP [RFC4656] and adheres to OWAMP's overall architecture and design. The TWAMP- Control and TWAMP-Test protocols accomplish their testing tasks as outlined below: - The Control-Client initiates a TCP connection on TWAMP's well- known port, and the Server (its role now established) responds with its Greeting message, indicating the security/integrity mode(s) it is willing to support. - The Control-Client responds with the chosen mode of communication and information supporting integrity protection and encryption, if the mode requires them. The Server responds to accept the mode and give its start time. This completes the control-connection setup. - The Control-Client requests (and describes) a test session with a unique TWAMP-Control message. The Server responds with its acceptance and supporting information. More than one test session may be requested with additional messages. - The Control-Client initiates all requested testing with a Start- Sessions message, and the Server acknowledges. - The Session-Sender and the Session-Reflector exchange test packets according to the TWAMP-Test protocol for each active session. - When appropriate, the Control-Client sends a message to stop all test sessions. There are two recognized extension mechanisms in the TWAMP Protocol. 1) The Modes field is used to establish the communication options during TWAMP-Control Connection Setup. 2) The TWAMP-Control Command Number is another intended extension mechanism, allowing additional commands to be defined in the future. The TWAMP-Control protocol resolves different capability levels between the Control-Client and Server. All multi-octet quantities defined in this document are represented as unsigned integers in network byte order, unless specified otherwise. Hedayat, et al. Standards Track [Page 5] RFC 5357 Two-Way Active Measurement Protocol October 2008 Throughout this memo, the bits marked MBZ (Must Be Zero) MUST be set to zero by senders and MUST be ignored by receivers. 3. TWAMP-Control TWAMP-Control is a derivative of the OWAMP-Control for two-way measurements. All TWAMP-Control messages are similar in format and follow similar guidelines to those defined in Section 3 of OWAMP [RFC4656] with the exceptions outlined in the following sections. One such exception is the Fetch-Session command, which is not used in TWAMP. 3.1. Connection Setup Connection establishment of TWAMP follows the same procedure defined in Section 3.1 of OWAMP [RFC4656]. The Modes field is a recognized extension mechanism in TWAMP, and the current mode values are identical to those used in OWAMP. The only exception is the well- known port number for TWAMP-Control. A Client opens a TCP connection to the Server on well-known port 862. The host that initiates the TCP connection takes the roles of Control-Client and (in the two-host implementation) the Session-Sender. The host that acknowledges the TCP connection accepts the roles of Server and (in the two-host implementation) the Session-Reflector. The Control-Client MAY set a desired code point in the Diffserv Code Point (DSCP) field in the IP header for ALL packets of a specific control connection. The Server SHOULD use the DSCP of the Control- Client's TCP SYN in ALL subsequent packets on that connection (avoiding any ambiguity in case of re-marking). The possibility exists for Control-Client failure after TWAMP- Control connection establishment, or the path between the Control- Client and Server may fail while a connection is in progress. The Server MAY discontinue any established control connection when no packet associated with that connection has been received within SERVWAIT seconds. The Server SHALL suspend monitoring control connection activity after receiving a Start-Sessions command, and SHALL resume after receiving a Stop-Sessions command (IF the SERVWAIT option is supported). Note that the REFWAIT timeout (described below) covers failures during test sessions, and if REFWAIT expires on ALL test sessions initiated by a TWAMP-Control connection, then the SERVWAIT monitoring SHALL resume (as though a Stop-Sessions command had been received). An implementation that supports the SERVWAIT timeout SHOULD also implement the REFWAIT timeout. The default value of SERVWAIT SHALL be 900 seconds, and this waiting time MAY be configurable. This timeout allows the Server to free up resources in case of failure. Hedayat, et al. Standards Track [Page 6] RFC 5357 Two-Way Active Measurement Protocol October 2008 Both the Server and the Client use the same mappings from KeyIDs to shared secrets. The Server, being prepared to conduct sessions with more than one Client, uses KeyIDs to choose the appropriate secret key; a Client would typically have different secret keys for different Servers. The shared secret is a passphrase. To maximize passphrase interoperability, the passphrase character set MUST be encoded using Appendix B of [RFC5198] (the ASCII Network Virtual Terminal Definition). It MUST not contain newlines (any combination of Carriage-Return (CR) and/or Line-Feed (LF) characters), and control characters SHOULD be avoided. 3.2. Integrity Protection Integrity protection of TWAMP follows the same procedure defined in Section 3.2 of OWAMP [RFC4656]. As in OWAMP, each HMAC (Hashed Message Authentication Code) sent covers everything sent in a given direction between the previous HMAC (but not including it) and the start of the new HMAC. This way, once encryption is set up, each bit of the TWAMP-Control connection is authenticated by an HMAC exactly once. Note that the Server-Start message (sent by a Server during the initial control-connection exchanges) does not terminate with an HMAC field. Therefore, the HMAC in the first Accept-Session message also covers the Server-Start message and includes the Start-Time field in the HMAC calculation. Also, in authenticated and encrypted modes, the HMAC in TWAMP-Control packets is encrypted. 3.3. Values of the Accept Field Accept values used in TWAMP are the same as the values defined in Section 3.3 of OWAMP [RFC4656]. 3.4. TWAMP-Control Commands TWAMP-Control commands conform to the rules defined in Section 3.4 of OWAMP [RFC4656]. The following commands are available for the Control-Client: Request-TW-Session, Start-Sessions, and Stop-Sessions. The Server can send specific messages in response to the commands it receives (as described in the sections that follow). Note that the OWAMP Request-Session command is replaced by the TWAMP Request-TW-Session command, and the Fetch-Session command does not appear in TWAMP. Hedayat, et al. Standards Track [Page 7] RFC 5357 Two-Way Active Measurement Protocol October 2008 3.5. Creating Test Sessions Test session creation follows the same procedure as defined in Section 3.5 of OWAMP [RFC4656]. The Request-TW-Session command is based on the OWAMP Request-Session command, and uses the message format as described in Section 3.5 of OWAMP, but without the Schedule Slot Descriptions field(s) and uses only one HMAC. The description of the Request-TW-Session format follows. In TWAMP, the first octet is referred to as the Command Number, and the Command Number is a recognized extension mechanism. Readers are encouraged to consult the TWAMP-Control Command Number registry to determine if there have been additional values assigned. The Command Number value of 5 indicates a Request-TW-Session command, and the Server MUST interpret this command as a request for a two-way test session using the TWAMP-Test protocol. If a TWAMP Server receives an unexpected Command Number, it MUST respond with the Accept field set to 3 (meaning "Some aspect of request is not supported") in the Accept-Session message. Command Numbers that are Forbidden (and possibly numbers that are Reserved) are unexpected. In OWAMP, the Conf-Sender field is set to 1 when the Request-Session message describes a task where the Server will configure a one-way test packet sender. Likewise, the Conf-Receiver field is set to 1 when the message describes the configuration for a Session-Receiver. In TWAMP, both endpoints send and receive test packets, with the Session-Sender first sending and then receiving test packets, complimented by the Session-Reflector first receiving and then sending. Both the Conf-Sender field and Conf-Receiver field MUST be set to 0 since the Session-Reflector will both receive and send packets, and the roles are established according to which host initiates the TCP connection for control. The Server MUST interpret any non-zero value as an improperly formatted command, and MUST respond with the Accept field set to 3 (meaning "Some aspect of request is not supported") in the Accept-Session message. The Session-Reflector in TWAMP does not process incoming test packets for performance metrics and consequently does not need to know the number of incoming packets and their timing schedule. Consequently the Number of Scheduled Slots and Number of Packets MUST be set to 0. The Sender Port is the UDP port from which TWAMP-Test packets will be sent and the port to which TWAMP-Test packets will be sent by the Hedayat, et al. Standards Track [Page 8] RFC 5357 Two-Way Active Measurement Protocol October 2008 Session-Reflector (the Session-Sender will use the same UDP port to send and receive packets). The Receiver Port is the desired UDP port to which TWAMP-Test packets will be sent by the Session-Sender (the port where the Session-Reflector is asked to receive test packets). The Receiver Port is also the UDP port from which TWAMP-Test packets will be sent by the Session-Reflector (the Session-Reflector will use the same UDP port to send and receive packets). The Sender Address and Receiver Address fields contain, respectively, the sender and receiver addresses of the endpoints of the Internet path over which a TWAMP-Test session is requested. They MAY be set to 0, in which case the IP addresses used for the Control-Client to Server TWAMP-Control message exchange MUST be used in the test packets. The Session Identifier (SID) is as defined in OWAMP [RFC4656]. Since the SID is always generated by the receiving side, the Server determines the SID, and the SID in the Request-TW-Session message MUST be set to 0. The Start Time is as defined in OWAMP [RFC4656]. The Timeout is interpreted differently from the definition in OWAMP [RFC4656]. In TWAMP, Timeout is the interval that the Session- Reflector MUST wait after receiving a Stop-Sessions message. In case there are test packets still in transit, the Session-Reflector MUST reflect them if they arrive within the Timeout interval following the reception of the Stop-Sessions message. The Session-Reflector MUST NOT reflect packets that are received beyond the timeout. Type-P descriptor is as defined in OWAMP [RFC4656]. The only capability of this field is to set the Differentiated Services Code Point (DSCP) as defined in [RFC2474]. The same value of DSCP MUST be used in test packets reflected by the Session-Reflector. Since there are no Schedule Slot Descriptions, the Request-TW-Session message is completed by MBZ (Must Be Zero) and HMAC fields. This completes one logical message, referred to as the Request-TW-Session command. The Session-Reflector MUST respond to each Request-TW-Session command with an Accept-Session message as defined in OWAMP [RFC4656]. When the Accept field = 0, the Port field confirms (repeats) the port to which TWAMP-Test packets are sent by the Session-Sender toward the Session-Reflector. In other words, the Port field indicates the port number where the Session-Reflector expects to receive packets from the Session-Sender. Hedayat, et al. Standards Track [Page 9] RFC 5357 Two-Way Active Measurement Protocol October 2008 When the requested Receiver Port is not available (e.g., port in use), the Server at the Session-Reflector MAY suggest an alternate and available port for this session in the Port field. The Session- Sender either accepts the alternate port, or composes a new Session- Request message with suitable parameters. Otherwise, the Server at the Control-Client uses the Accept field to convey other forms of session rejection or failure and MUST NOT suggest an alternate port; in this case, the Port field MUST be set to zero. 3.6. Send Schedules The send schedule for test packets defined in Section 3.6 of OWAMP [RFC4656] is not used in TWAMP. The Control-Client and Session- Sender MAY autonomously decide the send schedule. The Session- Reflector SHOULD return each test packet to the Session-Sender as quickly as possible. 3.7. Starting Test Sessions The procedure and guidelines for starting test sessions is the same as defined in Section 3.7 of OWAMP [RFC4656]. 3.8. Stop-Sessions The procedure and guidelines for stopping test sessions is similar to that defined in Section 3.8 of OWAMP [RFC4656]. The Stop-Sessions command can only be issued by the Control-Client. The message MUST NOT contain any session description records or skip ranges. The message is terminated with a single block HMAC to complete the Stop- Sessions command. Since the TWAMP Stop-Sessions command does not convey SIDs, it applies to all sessions previously requested and started with a Start-Sessions command. Hedayat, et al. Standards Track [Page 10] RFC 5357 Two-Way Active Measurement Protocol October 2008 Thus, the TWAMP Stop-Sessions command is constructed as follows: 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | 3 | Accept | MBZ | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Number of Sessions | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | MBZ (8 octets) | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | | | HMAC (16 octets) | | | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Above, the Command Number in the first octet (3) indicates that this is the Stop-Sessions command. Non-zero Accept values indicate a failure of some sort. Zero values indicate normal (but possibly premature) completion. The full list of available Accept values is described in Section 3.3 of [RFC4656], "Values of the Accept Field". If Accept has a non-zero value, results of all TWAMP-Test sessions spawned by this TWAMP-Control session SHOULD be considered invalid. If the Accept-Session message was not transmitted at all (for whatever reason, including failure of the TCP connection used for TWAMP-Control), the results of all TWAMP-Test sessions spawned by this TWAMP-Control session MAY be considered invalid. Number of Sessions indicates the number of sessions that the Control-Client intends to stop. Number of Sessions MUST contain the number of send sessions started by the Control-Client that have not been previously terminated by a Stop-Sessions command (i.e., the Control-Client MUST account for each accepted Request-Session). If the Stop-Sessions message does not account for exactly the number of sessions in progress, then it is to be considered invalid, the TWAMP-Control connection SHOULD be closed, and any results obtained considered invalid. Upon receipt of a TWAMP-Control Stop-Sessions command, the Session- Reflector MUST discard any TWAMP-Test packets that arrive at the current time plus the Timeout (in the Request-TW-Session command). Hedayat, et al. Standards Track [Page 11] RFC 5357 Two-Way Active Measurement Protocol October 2008 3.9. Fetch-Session One purpose of TWAMP is measurement of two-way metrics. Two-way measurement methods do not require packet-level data to be collected by the Session-Reflector (such as sequence number, timestamp, and Time to Live (TTL)) because this data is communicated in the "reflected" test packets. As such, the protocol does not require the retrieval of packet-level data from the Server and the OWAMP Fetch- Session command is not used in TWAMP. 4. TWAMP-Test The TWAMP-Test protocol is similar to the OWAMP-test protocol [RFC4656] with the exception that the Session-Reflector transmits test packets to the Session-Sender in response to each test packet it receives. TWAMP defines two different test packet formats, one for packets transmitted by the Session-Sender and one for packets transmitted by the Session-Reflector. As with OWAMP-test protocol [RFC4656], there are three modes: unauthenticated, authenticated, and encrypted. 4.1. Sender Behavior The sender behavior is determined by the configuration of the Session-Sender and is not defined in this standard. Further, the Session-Reflector does not need to know the Session-Sender behavior to the degree of detail as needed in OWAMP [RFC4656]. Additionally, the Session-Sender collects and records the necessary information provided from the packets transmitted by the Session-Reflector for measuring two-way metrics. The information recording based on the packet(s) received by the Session-Sender is implementation dependent. 4.1.1. Packet Timings Since the send schedule is not communicated to the Session-Reflector, there is no need for a standardized computation of packet timing. Regardless of any scheduling delays, each packet that is actually sent MUST have the best possible approximation of its real time of departure as its timestamp (in the packet). 4.1.2. Packet Format and Content The Session-Sender packet format and content follow the same procedure and guidelines as defined in Section 4.1.2 of OWAMP [RFC4656] (with the exception of the reference to the send schedule). Hedayat, et al. Standards Track [Page 12] RFC 5357 Two-Way Active Measurement Protocol October 2008 Note that the Reflector test packet formats are larger than the Sender's formats. The Session-Sender MAY append sufficient Packet Padding to allow the same IP packet payload lengths to be used in each direction of transmission (this is usually desirable). To compensate for the Reflector's larger test packet format, the Sender appends at least 27 octets of padding in unauthenticated mode, and at least 56 octets in authenticated and encrypted modes. 4.2. Reflector Behavior TWAMP requires the Session-Reflector to transmit a packet to the Session-Sender in response to each packet it receives. As packets are received, the Session-Reflector will do the following: - Timestamp the received packet. Each packet that is actually received MUST have the best possible approximation of its real time of arrival entered as its Received Timestamp (in the packet). - In authenticated or encrypted mode, decrypt the appropriate sections of the packet body (first block (16 octets) for authenticated, 96 octets for encrypted), and then check integrity of sections covered by the HMAC. - Copy the packet sequence number into the corresponding reflected packet to the Session-Sender. - Extract the Sender TTL value from the TTL/Hop Limit value of received packets. Session-Reflector implementations SHOULD fetch the TTL/Hop Limit value from the IP header of the packet, replacing the value of 255 set by the Session-Sender. If an implementation does not fetch the actual TTL value (the only good reason not to do so is an inability to access the TTL field of arriving packets), it MUST set the Sender TTL value as 255. - In authenticated and encrypted modes, the HMAC MUST be calculated first, then the appropriate portion of the packet body is encrypted. - Transmit a test packet to the Session-Sender in response to every received packet. The response MUST be generated as immediately as possible. The format and content of the test packet is defined in Section 4.2.1. Prior to the transmission of the test packet, the Session-Reflector MUST enter the best possible approximation of its actual sending time as its Timestamp (in the packet). This permits the determination of the elapsed time between the reception of the packet and its transmission. Hedayat, et al. Standards Track [Page 13] RFC 5357 Two-Way Active Measurement Protocol October 2008 - Packets not received within the Timeout (following the Stop- Sessions command) MUST be ignored by the Reflector. The Session- Reflector MUST NOT generate a test packet to the Session-Sender for packets that are ignored. The possibility exists for Session-Sender failure during a session, or the path between the Session-Sender and Session-Reflector may fail while a test session is in progress. The Session-Reflector MAY discontinue any session that has been started when no packet associated with that session has been received for REFWAIT seconds. The default value of REFWAIT SHALL be 900 seconds, and this waiting time MAY be configurable. This timeout allows a Session-Reflector to free up resources in case of failure. 4.2.1. TWAMP-Test Packet Format and Content The Session-Reflector MUST transmit a packet to the Session-Sender in response to each packet received. The Session-Reflector SHOULD transmit the packets as immediately as possible. The Session- Reflector SHOULD set the TTL in IPv4 (or Hop Limit in IPv6) in the UDP packet to 255. The test packet will have the necessary information for calculating two-way metrics by the Session-Sender. The format of the test packet depends on the mode being used. The two formats are presented below. Hedayat, et al. Standards Track [Page 14] RFC 5357 Two-Way Active Measurement Protocol October 2008 For unauthenticated mode: 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Sequence Number | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Timestamp | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Error Estimate | MBZ | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Receive Timestamp | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Sender Sequence Number | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Sender Timestamp | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Sender Error Estimate | MBZ | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Sender TTL | | +-+-+-+-+-+-+-+-+ + | | . . . Packet Padding . . . | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Hedayat, et al. Standards Track [Page 15] RFC 5357 Two-Way Active Measurement Protocol October 2008 For authenticated and encrypted modes: 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Sequence Number | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | MBZ (12 octets) | | | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Timestamp | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Error Estimate | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ + | MBZ (6 octets) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Receive Timestamp | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | MBZ (8 octets) | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Sender Sequence Number | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | MBZ (12 octets) | | | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Sender Timestamp | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Sender Error Estimate | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ + | MBZ (6 octets) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Sender TTL | | +-+-+-+-+-+-+-+-+ + | | | | | MBZ (15 octets) | +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ | HMAC (16 octets) | | | | | | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-| Hedayat, et al. Standards Track [Page 16] RFC 5357 Two-Way Active Measurement Protocol October 2008 | | . . . Packet Padding . . . | | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Note that all timestamps have the same format as OWAMP [RFC4656] as follows: 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Integer part of seconds | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Fractional part of seconds | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Sequence Number is the sequence number of the test packet according to its transmit order. It starts with zero and is incremented by one for each subsequent packet. The Sequence Number generated by the Session-Reflector is independent from the sequence number of the arriving packets. Timestamp and Error Estimate are the Session-Reflector's transmit timestamp and error estimate for the reflected test packet, respectively. The format of all timestamp and error estimate fields follow the definition and formats defined by OWAMP, Section 4.1.2 in [RFC4656]. Sender Timestamp and Sender Error Estimate are exact copies of the timestamp and error estimate from the Session-Sender test packet that corresponds to this test packet. Sender TTL is 255 when transmitted by the Session-Sender. Sender TTL is set to the Time To Live (or Hop Count) value of the received packet from the IP packet header when transmitted by the Session- Reflector. Receive Timestamp is the time the test packet was received by the reflector. The difference between Timestamp and Receive Timestamp is the amount of time the packet was in transition in the Session- Reflector. The Error Estimate associated with the Timestamp field also applies to the Receive Timestamp. Sender Sequence Number is a copy of the Sequence Number of the packet transmitted by the Session-Sender that caused the Session-Reflector to generate and send this test packet. Hedayat, et al. Standards Track [Page 17] RFC 5357 Two-Way Active Measurement Protocol October 2008 The HMAC field in TWAMP-Test packets covers the same fields as the Advanced Encryption Standard (AES) encryption. Thus, in authenticated mode, HMAC covers the first block (16 octets); in encrypted mode, HMAC covers the first six blocks (96 octets). In TWAMP-Test, the HMAC field MUST NOT be encrypted. Packet Padding in TWAMP-Test SHOULD be pseudo-random (it MUST be generated independently of any other pseudo-random numbers mentioned in this document). However, implementations MUST provide a configuration parameter, an option, or a different means of making Packet Padding consist of all zeros. Packet Padding MUST NOT be covered by the HMAC and MUST NOT be encrypted. The minimum data segment length of TWAMP-Test packets in unauthenticated mode is 41 octets, and 104 octets in both authenticated mode and encrypted modes. Note that the Session-Reflector Test packet formats are larger than the Sender's formats. The Session-Reflector SHOULD reduce the length of the Sender's Packet Padding to achieve equal IP packet payload lengths in each direction of transmission, when sufficient padding is present. The Session-Reflector MAY re-use the Sender's Packet Padding (since the requirements for padding generation are the same for each), and in this case the Session-Reflector SHOULD truncate the padding such that the highest-number octets are discarded. In unauthenticated mode, encryption or authentication MUST NOT be applied. The TWAMP-Test packet layout is identical in authenticated and encrypted modes. The encryption operation for a Session-Sender packet follows the same rules of Session-Sender packets as defined in OWAMP section 4.1.2 of [RFC4656]. The main difference between authenticated mode and encrypted mode is the portion of the test packets that are covered by HMAC and encrypted. Authenticated mode permits the timestamp to be fetched after a portion of the packet is encrypted, but in encrypted mode all the sequence numbers and timestamps are fetched before encryption to provide maximum data-integrity protection. In authenticated mode, only the sequence number in the first block is encrypted, and the subsequent timestamps and sequence numbers are sent in clear text. Sending the timestamp in clear text allows one to reduce the time between when a timestamp is obtained by a Session-Reflector and when that packet is sent out. This potentially improves the timestamp accuracy, because the sequence number can be encrypted before the timestamp is fetched. Hedayat, et al. Standards Track [Page 18] RFC 5357 Two-Way Active Measurement Protocol October 2008 In encrypted mode, the reflector MUST fetch the timestamps, generate the HMAC, and encrypt the packet, then send it. Obtaining the keys and encryption methods follows the same procedure as OWAMP as described below. Each TWAMP-Test session has two keys, an AES Session-key and an HMAC Session-key, and the keys are derived from the TWAMP-Control keys and the SID. The TWAMP-Test AES Session-key is obtained as follows: the TWAMP- Control AES Session-key (the same AES Session-key as used for the corresponding TWAMP-Control session) is encrypted with the 16-octet session identifier (SID) as the key, using a single-block AES-ECB encryption. The result is the TWAMP-Test AES Session-key to be used in encrypting (and decrypting) the packets of the particular TWAMP- Test session. Note that the TWAMP-Test AES Session-key, TWAMP- Control AES Session-key, and the SID are all comprised of 16 octets. The TWAMP-Test HMAC Session-key is obtained as follows: the TWAMP- Control HMAC Session-key (the same HMAC Session-key as used for the corresponding TWAMP-Control session) is encrypted using AES-CBC (Cipher Block Chaining) with the 16-octet session identifier (SID) as the key. This is a two-block CBC encryption that is always performed with IV=0. Note that the TWAMP-Test HMAC Session-key and TWAMP- Control HMAC Session-key are comprised of 32 octets, while the SID is 16 octets. In authenticated mode, the first block (16 octets) of each TWAMP-Test packet is encrypted using the AES Electronic Codebook (ECB) mode. This mode does not involve any chaining, and lost, duplicated, or reordered packets do not cause problems with deciphering any packet in a TWAMP-Test session. In encrypted mode, the first six blocks (96 octets) are encrypted using AES-CBC mode. The AES Session-key to use is obtained in the same way as the key for authenticated mode. Each TWAMP-Test packet is encrypted as a separate stream, with just one chaining operation; chaining does not span multiple packets so that lost, duplicated, or reordered packets do not cause problems. The initialization vector for the CBC encryption is a value with all bits equal to zero. Implementation Note: Naturally, the key schedule for each TWAMP-Test session MUST be set up at most once per session, not once per packet. Hedayat, et al. Standards Track [Page 19] RFC 5357 Two-Way Active Measurement Protocol October 2008 5. Implementers' Guide This section serves as guidance to implementers of TWAMP. The example architecture presented here is not a requirement. Similar to OWAMP [RFC4656], TWAMP is designed with enough flexibility to allow different architectures that suit multiple system requirements. In this example, the roles of Control-Client and Session-Sender are implemented in one host referred to as the controller, and the roles of Server and Session-Reflector are implemented in another host referred to as the responder. controller responder +-----------------+ +-------------------+ | Control-Client |<--TWAMP-Control-->| Server | | | | | | Session-Sender |<--TWAMP-Test----->| Session-Reflector | +-----------------+ +-------------------+ This example provides an architecture that supports the full TWAMP standard. The controller establishes the test session with the responder through the TWAMP-Control protocol. After the session is established, the controller transmits test packets to the responder. The responder follows the Session-Reflector behavior of TWAMP as described in Section 4.2. Appendix I provides an example for purely informational purposes. It suggests an incremental path to adopting TWAMP, by implementing the TWAMP-Test protocol first. 6. Security Considerations Fundamentally, TWAMP and OWAMP use the same protocol for establishment of Control and Test procedures. The main difference between TWAMP and OWAMP is the Session-Reflector behavior in TWAMP vs. the Session-Receiver behavior in OWAMP. This difference in behavior does not introduce any known security vulnerabilities that are not already addressed by the security features of OWAMP. The entire security considerations of OWAMP [RFC4656] applies to TWAMP. The Server-Greeting message (defined in OWAMP, Section 3.1 of [RFC4656]) includes a Count field to specify the iteration counter used in PKCS #5 to generate keys from shared secrets. OWAMP recommends a lower limit of 1024 iterations, but no upper limit. The Count field provides an opportunity for a denial-of-service (DOS) attack because it is 32 bits long. If an attacking system set the maximum value in Count (2**32), then the system under attack would stall for a significant period of time while it attempts to generate Hedayat, et al. Standards Track [Page 20] RFC 5357 Two-Way Active Measurement Protocol October 2008 keys. Therefore, TWAMP-compliant systems SHOULD have a configuration control to limit the maximum Count value. The default maximum Count value SHOULD be 32768. As suggested in OWAMP, this value MAY be increased when greater computing power becomes common. If a Control-Client receives a Server-Greeting message with Count greater that its maximum configured value, it SHOULD close the control connection. 7. Acknowledgements We would like to thank Nagarjuna Venna, Sharee McNab, Nick Kinraid, Stanislav Shalunov, Matt Zekauskas, Walt Steverson, Jeff Boote, Murtaza Chiba, and Kevin Earnst for their comments, suggestions, reviews, helpful discussion, and proof-reading. Lars Eggert, Sam Hartman, and Tim Polk contributed very useful AD-level reviews, and the authors thank them for their contributions to the memo. 8. IANA Considerations IANA has allocated a well-known TCP port number (861) for the OWAMP- Control part of the OWAMP [RFC4656] protocol. ... owamp-control 861/tcp OWAMP-Control owamp-control 861/udp OWAMP-Control # [RFC4656] IANA has also allocated a well-known TCP/UDP port number for the TWAMP-Control protocol. ... twamp-control 862/tcp Two-way Active Measurement Protocol (TWAMP) Control twamp-control 862/udp Two-way Active Measurement Protocol (TWAMP) Control # [RFC5357] # 863-872 Unassigned Since TWAMP adds an additional Control command beyond the OWAMP- Control specification and describes behavior when this control command is used, IANA has created a registry for the TWAMP Command Number field. The field is not explicitly named in [RFC4656] but is called out for each command. This field is a recognized extension mechanism for TWAMP. Hedayat, et al. Standards Track [Page 21] RFC 5357 Two-Way Active Measurement Protocol October 2008 8.1. Registry Specification IANA has created a TWAMP-Control Command Number registry. TWAMP- Control commands are specified by the first octet in OWAMP-Control messages as shown in Section 3.5 of [RFC4656], and modified by this document. Thus, this registry may contain sixteen possible values. 8.2. Registry Management Because the registry may only contain sixteen values, and because OWAMP and TWAMP are IETF protocols, this registry must only be updated by "IETF Consensus" as specified in [RFC5226] -- an RFC documenting the use that is approved by the IESG. We expect that new values will be assigned as monotonically increasing integers in the range [0-15], unless there is a good reason to do otherwise. 8.3. Experimental Numbers [RFC3692] recommends allocating an appropriate number of values for experimentation and testing. It is not clear to the authors exactly how many numbers might be useful in this space, or if it would be useful that they were easily distinguishable or at the "high end" of the number range. Two might be useful, say one for session control, and one for session fetch. On the other hand, a single number would allow for unlimited extension, because the format of the rest of the message could be tailored, with allocation of other numbers done once usefulness has been proven. Thus, this document allocates one number (6) as designated for experimentation and testing. 8.4. Initial Registry Contents TWAMP-Control Command Number Registry Value Description Semantics Definition 0 Reserved 1 Forbidden 2 Start-Sessions RFC 4656, Section 3.7 3 Stop-Sessions RFC 4656, Section 3.8 4 Reserved 5 Request-TW-Session this document, Section 3.5 6 Experimentation undefined, see Section 8.3. 9. Internationalization Considerations The protocol does not carry any information in a natural language, with the possible exception of the KeyID in TWAMP-Control, which is encoded in UTF-8 [RFC3629, RFC5198]. Hedayat, et al. Standards Track [Page 22] RFC 5357 Two-Way Active Measurement Protocol October 2008 Appendix I - TWAMP Light (Informative) In this example, the roles of Control-Client, Server, and Session- Sender are implemented in one host referred to as the controller, and the role of Session-Reflector is implemented in another host referred to as the responder. controller responder +-----------------+ +-------------------+ | Server |<----------------->| | | Control-Client | | Session-Reflector | | Session-Sender |<--TWAMP-Test----->| | +-----------------+ +-------------------+ This example provides a simple architecture for responders where their role will be to simply act as light test points in the network. The controller establishes the test session with the Server through non-standard means. After the session is established, the controller transmits test packets to the responder. The responder follows the Session-Reflector behavior of TWAMP as described in section 4.2 with the following exceptions. In the case of TWAMP Light, the Session-Reflector does not necessarily have knowledge of the session state. IF the Session- Reflector does not have knowledge of the session state, THEN the Session-Reflector MUST copy the Sequence Number of the received packet to the Sequence Number field of the reflected packet. The controller receives the reflected test packets and collects two-way metrics. This architecture allows for collection of two-way metrics. This example eliminates the need for the TWAMP-Control protocol, and assumes that the Session-Reflector is configured and communicates its configuration with the Server through non-standard means. The Session-Reflector simply reflects the incoming packets back to the controller while copying the necessary information and generating sequence number and timestamp values per Section 4.2.1. TWAMP Light introduces some additional security considerations. The non-standard means to control the responder and establish test sessions SHOULD offer the features listed below. The non-standard responder control protocol SHOULD have an authenticated mode of operation. The responder SHOULD be configurable to accept only authenticated control sessions. The non-standard responder control protocol SHOULD have a means to activate the authenticated and encrypted modes of the TWAMP-Test protocol. Hedayat, et al. Standards Track [Page 23] RFC 5357 Two-Way Active Measurement Protocol October 2008 When the TWAMP Light test sessions operate in authenticated or encrypted mode, the Session-Reflector MUST have some mechanism for generating keys (because the TWAMP-Control protocol normally plays a role in this process, but is not present here). The specification of the key generation mechanism is beyond the scope of this memo. Normative References [RFC4656] Shalunov, S., Teitelbaum, B., Karp, A., Boote, J., and M. Zekauskas, "A One-way Active Measurement Protocol (OWAMP)", RFC 4656, September 2006. [RFC2681] Almes, G., Kalidindi, S., and M. Zekauskas, "A Round-trip Delay Metric for IPPM", RFC 2681, September 1999. [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. [RFC2474] Nichols, K., Blake, S., Baker, F., and D. Black, "Definition of the Differentiated Services Field (DS Field) in the IPv4 and IPv6 Headers", RFC 2474, December 1998. [RFC5226] Narten, T. and H. Alvestrand, "Guidelines for Writing an IANA Considerations Section in RFCs", BCP 26, RFC 5226, May 2008. [RFC3629] Yergeau, F., "UTF-8, a transformation format of ISO 10646", STD 63, RFC 3629, November 2003. [RFC5198] Klensin, J. and M. Padlipsky, "Unicode Format for Network Interchange", RFC 5198, March 2008. Informative References [RFC3692] Narten, T., "Assigning Experimental and Testing Numbers Considered Useful", BCP 82, RFC 3692, January 2004. Hedayat, et al. Standards Track [Page 24] RFC 5357 Two-Way Active Measurement Protocol October 2008 Authors' Addresses Kaynam Hedayat Brix Networks 285 Mill Road Chelmsford, MA 01824 USA EMail: khedayat@brixnet.com URI: http://www.brixnet.com/ Roman M. Krzanowski, Ph.D. Verizon 500 Westchester Ave. White Plains, NY USA EMail: roman.krzanowski@verizon.com URI: http://www.verizon.com/ Al Morton AT&T Labs Room D3 - 3C06 200 Laurel Ave. South Middletown, NJ 07748 USA Phone +1 732 420 1571 EMail: acmorton@att.com URI: http://home.comcast.net/~acmacm/ Kiho Yum Juniper Networks 1194 Mathilda Ave. Sunnyvale, CA USA EMail: kyum@juniper.net URI: http://www.juniper.com/ Jozef Z. Babiarz Nortel Networks 3500 Carling Avenue Ottawa, Ont K2H 8E9 Canada Email: babiarz@nortel.com URI: http://www.nortel.com/ Hedayat, et al. Standards Track [Page 25] RFC 5357 Two-Way Active Measurement Protocol October 2008 Full Copyright Statement Copyright (C) The IETF Trust (2008). This document is subject to the rights, licenses and restrictions contained in BCP 78, and except as set forth therein, the authors retain all their rights. This document and the information contained herein are provided on an "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY, THE IETF TRUST AND THE INTERNET ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Intellectual Property The IETF takes no position regarding the validity or scope of any Intellectual Property Rights or other rights that might be claimed to pertain to the implementation or use of the technology described in this document or the extent to which any license under such rights might or might not be available; nor does it represent that it has made any independent effort to identify any such rights. Information on the procedures with respect to rights in RFC documents can be found in BCP 78 and BCP 79. Copies of IPR disclosures made to the IETF Secretariat and any assurances of licenses to be made available, or the result of an attempt made to obtain a general license or permission for the use of such proprietary rights by implementers or users of this specification can be obtained from the IETF on-line IPR repository at http://www.ietf.org/ipr. The IETF invites any interested party to bring to its attention any copyrights, patents or patent applications, or other proprietary rights that may cover technology that may be required to implement this standard. Please address the information to the IETF at ietf-ipr@ietf.org. Hedayat, et al. Standards Track [Page 26]