Edvina AB

Runbovägen 10 Sollentuna SE-192 48 SE oej@edvina.net

Cisco Systems

7200-12 Kit Creek Road Research Triangle Park NC 27709 US gsalguei@cisco.com

Ariadne Internet Services

738 Main St. Waltham MA 02451 US worley@ariadne.com

Applications and Real-Time Area (ART) SIPCORE I-D Internet-Draft The Session Initiation Protocol (SIP) supports multiple transports running both over IPv4 and IPv6 protocols. In more and more cases, a SIP user agent (UA) is connected to multiple network interfaces. In these cases setting up a connection from a dual stack client to a dual stack server may suffer from the issues described in RFC 6555 ("Happy Eyeballs") - significant delays in the process of setting up a working flow to a server. This negatively affects user experience. This document builds on RFC 6555 and explains how a compliant SIP implementation can minimize delays when contacting a host name (obtained by using DNS NAPTR and SRV lookups) in a dual stack network using connection-oriented transport protocols.

The Session Initiation Protocol (SIP) and the documents that extended it provide support for both IPv4 and IPv6. However, this support has problems with environments that are characteristic of the transitional migratory phase from IPv4 to IPv6 networks. During this phase, many server and client implementations run on dual-stack hosts. In such environments, a dual-stack host will likely suffer greater connection delay, and by extension an inferior user experience, than an IPv4-only host. The need to remedy this diminished performance of dual-stack hosts led to the development of the "Happy Eyeballs" algorithm, which has since been implemented in many protocols and applications. This document revises the the procedures to apply the "Happy Eyeballs" framework. A dual-stack client using connection-oriented transport should set up multiple connections in parallel, to targets based on the result of DNS queries. This document starts at the point where a SIP implementation has a host name that resolves using A and AAAA records. Such a host name can either be the host part of a SIP URI (possibly including a port number) or the result of a lookup using DNS NAPTR and SRV records as described in RFC 3263 (as updated by RFC 7984). Procedures for connectionless transport protocols for SIP are outside the scope of this document. Procedures allowing a client to change the order of contacting targets that were derived from different host names are outside the scope of this document. The concepts in this document are elaborated from those developed in , and so some background information in RFC 6555 is not repeated here. The reader is encouraged to read the available documentation regarding implementations of RFC 6555, as well as study Open Source implementations, in order to learn from the experience accumulated since the publishing of RFC 6555 in 2012.

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 . RFC 3261 defines additional terms used in this document that are specific to the SIP domain such as "proxy"; "registrar"; "redirect server"; "user agent server" or "UAS"; "user agent client" or "UAC"; "back-to-back user agent" or "B2BUA"; "dialog"; "transaction"; "server transaction". This document uses the term "SIP Server" that is defined to include the following SIP entities: user agent server, registrar, redirect server, a SIP proxy in the role of user agent server, and a B2BUA in the role of a user agent server. This document also uses the following terminology to make clear distinction between SIP entities supporting only IPv4, only IPv6 or supporting both IPv4 and IPv6. An IPv4-only UA/UAC/UAS supports SIP signaling and media only on the IPv4 network. It does not understand IPv6 addresses. An IPv6-only UA/UAC/UAS supports SIP signaling and media only on the IPv6 network. It does not understand IPv4 addresses. A UA/UAC/UAS that supports SIP signaling and media on both IPv4 and IPv6 networks; such a UA/UAC/UAS is known (and will be referred to in this document) as a "dual-stack" UA/UAC/UAS. Discussion: Do we need special handling of websocket transport? While this document uses the term "dual-stack" based on RFC 6555 and earlier terminology, the authors acknowledge that the same solution can be applied to multi-interface environments as well as future versions of IP alongside with the current ones.

A SIP client uses DNS to find a server based on a SIP URI. This process is described in and updated in . Using this process, a list of "targets" is constructed, where each target consists of an IP address, a port number, and a protocol (e.g., TCP, UDP, TLS) by which to contact that address/port. The process proceeds by constructing a sequence of host names, possibly by looking up NAPTR and/or SRV DNS records, and then for each host name looking up DNS address records (for all address families supported by the client) to generate the list of IP addresses for targets that are derived from that host name. The addresses for each host name are ordered using the client's destination selection rules. The sorted targets for all the host names are then concatenated into the sequence of targets to which the client will attempt to send the SIP message. Previously, the client contacts the targets in order until one is contacted successfully. In order to contact a target, the client establishes a transport connection (if necessary), sends the message using the transport (possibly resending the message several times), and then (for requests) waits for a response (either provisional or final). The process ends successfully if the client receives a response. The process ends unsuccessfully if the client receives a permanent error from the transport layer or if a SIP timer (Timer B or Timer F in ) expires. Timeouts generally default to 32 seconds. If the user has to wait for even one timeout, this will seriously degrade the user experience. Thus, it is desirable to minimize the number of times the client has timeouts when sending requests. If the target list contains both IPv6 addresses and IPv4 addresses, this procedure can degrade the user's experience in common situations. Typically, this problem arises when the client has an IPv6 interface, the server's preferred address is an IPv6 address, but the transit networks between the client and server do not carry IPv6. This can cause the client to attempt to send a SIP request for 32 seconds before it times out that target and continues with an IPv4 target. This problem parallels a problem that was widely seen in web browsers that was cured by specifying that web browsers should use a "Happy Eyeballs" algorithm to determine the order in which to contact target addresses. This document specifies an amendment to these procedures, by which the subsequences of targets derived from individual host names may be contacted in a different order than is specified by the destination selection rules. As in , the algorithm that the client uses is not specified by this document, but this document places requirements on the algorithm that improve the user's experience without unduly burdening the Internet infrastructure. By analogy with the name "Happy Eyeballs" for similar algorithms in web browsers, we label these algorithms "Happy EarBalls". This document modifies the transport procedures only in the case when all targets for a host name have connection-oriented protocols (currently, TCP and TLS). Other cases are outside the scope of this document.

This section discusses the situation that most closely resembles RFC 6555, which is when the SIP client has no active connection to any of the targets in a subsequence of targets derived from one host name. This specification allows the client to attempt to send a request to targets in the subsequence in a different order than is prescribed by RFC 3262 and RFC 6724. In addition, this specification allows the client to attempt to initiate a connection to a target without subsequently sending a request to the target. However, the algorithm which the client uses use meet the constraints in this section. Typically, the SIP client will set up two connections, with some head start for one address family (which is possibly be configurable) and then select the first completed connection for use and close the other one. The SIP message is sent on the selected connection only. The reason for this approach is to avoid the timeout associated with sending an unsuccessful SIP request, requiring the client to wait for a timeout before the request can be sent on a connection to another target - which in the case of SIP with default timers is 32 seconds. Waiting for timeout before trying with a second address will lead to a very poor user experience.

The following requirements apply to any implementation that takes advantage of the relaxed requirements on message transmission specified by this document.

An implementation MUST prefer the first IP address family returned by the host's address preference policy, unless it implements a stateful algorithm as described in . This usually means giving preference to IPv6 over IPv4, although that preference can be overridden by user configuration or by network configuration. If the host's policy is unknown or not attainable, the implementation MUST prefer IPv6 over IPv4.

The algorithm may be stateful -- that is, the algorithm will remember that IPv6 always fails, or that IPv6 to certain prefixes always fails, and so on. This section constrains such algorithms. Stateless algorithms, which do not remember the success/failure of previous connections, are not discussed in this section. After making a connection attempt using the preferred address family (e.g., IPv6) and failing to establish a connection within a certain time period (see ), a Happy EarBalls implementation will decide to initiate a second connection attempt using the same address family or the other address family. Such an implementation MAY make prioritize making subsequent connection attempts (to the same host or to other hosts) using the successful address family (e.g., IPv4). So long as new connections are being attempted by the host, such an implementation MUST occasionally make connection attempts using the host's preferred address family, as that family may have become functional again, and the client SHOULD do so every 10 minutes. The 10-minute delay before retrying a failed address family avoids the simple doubling of connection attempts on both IPv6 and IPv4. This can be achieved by flushing Happy EarBalls state every 10 minutes, which does not significantly harm the application's subsequent connection setup time. If connections using the preferred address family are again successful, the preferred address family MUST be used for subsequent connections. Because this implementation is stateful, it MAY track connection success (or failure) based on IPv6 or IPv4 prefix (e.g., connections to the same prefix assigned to the interface are successful whereas connections to other prefixes are failing).

Because every network has different characteristics (e.g., working or broken IPv6 or IPv4 connectivity), a Happy EarBalls algorithm SHOULD re-initialize when the interface is connected to a new network. Interfaces can determine network (re-)initialization by a variety of mechanisms (e.g., Detecting Network Attachment in IPv4 (DNAv4) [RFC4436], DNAv6 [RFC6059]).

Non-winning connections SHOULD be abandoned, even though they could -- in some cases -- be put to reasonable use. Justification: This reduces the load on the server (file descriptors, TCP control blocks) and stateful middleboxes (NAT and firewalls). Also, if the abandoned connection is IPv4, this reduces IPv4 address sharing contention. (There are some unlikely situations where a non-winning connection could be useful in the future: If at a later time, the client must send a request to a different host name, but one which has as a target the peer of the non-winning connection and does not have as a target the peer of the winning connection.)

When a client desires to send a message to a target that is within a subsequence of targets derived from one host name, the client may already have a connection established to one of the targets through either SIP Outbound or the procedures of . The client SHOULD attempt to send the message using the existing connection in preference to using a new connection to one of the targets. If, in the client's operational environment, there is a significant risk that the connection has become unusable without the client becoming aware of it, the client SHOULD consider testing whether the connection is usable before sending the message using the connection. Some possible ways to probe a connection to determine if it is still usable are: Send a keep-alive, as specified by the protocol of the connection. Send a CR-LF-CR-LF keep-alive on a SIP Outbound connection. Send an OPTIONS request with "Max-Forwards: 0". (Note that a probe using an OPTIONS request can be used with any protocol. If the OPTIONS reaches the target, the target is required to respond with either a 200 or 483 response without forwarding it to another entity. Conveniently, a server can respond to such a request statelessly, so such requests are low-overhead. (Although the keep-alive methods have even lower overhead.))

This section discusses additional considerations related to Happy EarBalls.

A client may be in a situation where it has advance notice that it is likely to need to send a message to a particular host name, for instance, if the user of a UA begins dialing an outgoing call which will be routed through a particular outgoing proxy. In such a situation, the client SHOULD consider preemptively establishing a connection () or probing an existing connection ().

For some transitional technologies, such as a dual-stack host, it is easy for the application to recognize a native IPv6 address (learned via a AAAA query) and a native IPv4 address (learned via an A query). The use of IPv6/IPv4 translation in the local network makes it difficult or impossible to determine the address family by which the connection will traverse the global network. However, IPv6/IPv4 translators do not need to be deployed on networks with dual-stack clients because dual-stack clients can use their native IP address family. Environments where IPv6/IPv4 translation is active will degrade the ability of Happy EarBalls algorithms to establish working connections.

Happy EarBalls is aimed at ensuring a reliable user experience regardless of connectivity problems affecting any single transport. However, this naturally means that applications employing these techniques are by default less useful for diagnosing issues with a particular address family. To assist in that regard, an implementation MAY provide a mechanism to disable their Happy EarBalls behavior via a user setting, and to provide data useful for debugging (e.g., a log or way to review current preferences).

A dual-stack host normally has two logical interfaces: an IPv6 interface and an IPv4 interface. However, a dual-stack host might have more than two logical interfaces because of a VPN (where a third interface is the tunnel address, often assigned by the remote corporate network), because of multiple physical interfaces such as wired and wireless Ethernet, because the host belongs to multiple VLANs, or other reasons. Optimal operation of Happy EarBalls with more than two logical interfaces is for further study and is outside the scope of this document.

It is possible that a DNS query for an A or AAAA resource record will return more than one A or AAAA address. When this occurs, it is RECOMMENDED that a Happy EarBalls implementation order the responses following the host's address preference policy and then try the first target. If that fails after a certain time (see ), the next target SHOULD be chosen from the other address family. If the second attempt fails to connect, a Happy EarBalls implementation SHOULD try the other targets; the order of these connection attempts is not important. On the Internet today, servers commonly have multiple A records to provide load-balancing across their servers. This same technique would be useful for AAAA records, as well. However, if multiple AAAA records are returned to a client that is not using Happy EarBalls and that has broken IPv6 connectivity, the multiple AAAA records will further increase the delay to fall back to IPv4. Thus, SIP server operators with native IPv6 connectivity SHOULD NOT offer multiple AAAA records. If Happy EarBalls is widely deployed in the future, this recommendation might be revisited.

The primary purpose of Happy EarBalls is to reduce the wait time for a dual-stack connection to complete, especially when the IPv6 path is broken and IPv6 is preferred. Using a short timeout between initiating an IPv6 connection and initiating an IPv4 connection (on the order of tens of milliseconds) achieves this goal, but at the cost of network traffic. This network traffic may be billable on certain networks, will create state on some middleboxes (e.g., firewalls, intrusion detection systems, NATs), and will consume ports if IPv4 addresses are shared. For these reasons, it is RECOMMENDED that connection attempts be paced to give connections a chance to complete. It is RECOMMENDED that connection attempts be paced 150-250 ms apart to balance human factors against network load. A stateful algorithm MAY be more aggressive (that is, make connection attempts closer together), if it maintains estimates of the expected connection completion times.

This document places additional restrictions on the existing procedures in the SIP protocol. The specific security vulnerabilities, attacks and threat models of the various protocols discussed in this document (SIP, DNS, SRV records, etc.) are well-documented in their respective specifications.

This document does not require any actions by IANA.

Note to RFC Editor: Upon publication, remove this section.

This version has a different name for technical reasons. It is, in reality, the successor to draft-johansson-sip-he-connection-01. Move Acknowledgments after References, as that is the style the Editor prefers. Updated Security Considerations: This increment of the H.E. work does not make normative changes in existing SIP. Copy a lot of text from RFC 6555, as this I-D is parallel to RFC 6555. Changed "hostname" to "host name", as the latter form is more common in RFCs by a moderate margin. Revised some of the introduction text to parallel the introduction of RFC 7984. Changed name of algorithm to "Happy EarBalls", added reference to Urban Dictionary. Many expansions of the discussion and revisions of the wording.

The authors would like to acknowledge the support and contribution of the SIP Forum IPv6 Working Group. This document is based on a lot of tests and discussions at SIPit events, organized by the SIP Forum. Most of the material in is taken from , whose authors are Dan Wing and Andrew Yourtchenko.