SIMPLE Buddylist Configuration Problem Statement

One of the formal deliverables of the SIMPLE working group is to provide an architecture that allows multiple interoperable implementations to provide a traditional buddylist-based instant messaging presence application using SIP. An informal design goal of the working group that stems from this deliverable is that such solutions should enable at least the same set of features as the currently available proprietary offerings. One of the keystones in realizing this goal is allowing developers to provide a user experience that is as good as or better than such offerings. One stumbling block in allowing developers to create such a user experience is the fact that there is currently no defined way, given a user's address of record, to retrieve a list of contacts for the purposes of displaying presence data and conveniently sending messages. Without such an ability, creating a user experience that is as straightforward as those currently available is frustrated.

Imagine a typical internet user, known for the purposes of this description as Bob, who wants to walk into a random Internet cafe and log into an IM client so that he can see who of his friends are online, and begin to send and receive messages. With currently deployed proprietary systems, Bob would be able to fire up the client, type in his user name and his password, and be finished. With no further interaction, Bob's presence information is changed, servers know how to route incoming messages to Bob, Bob's buddylist is displayed to him, and client starts receiving updates which indicate which of his buddies are online. The underlying proprietary protocol knows, given a user name of, e.g., "bob1963", how to perform all of these actions. SIMPLE currently has a hole in this area. Client creators can acheive almost all of the effects described above using mechanisms already defined or under development within the IETF. Assuming that Bob remembers his user ID (sip:bob@example.com, which is nicely mnemonic and probably matches one of his e-mail addresses) and password (used for responses to digest challenges), the client can send a REGISTER to sip:example.com (to route messages to him), send a PUBLISH to sip:bob@example.com (to update his presence), and send an event-list aware presence SUBSCRIBE to get his buddy list and the status of each buddy. The complication arises from the fact that the client doesn't have a URI to which this SUBSCRIBE can be sent. So, without prompting Bob for an additional URI -- that of his buddy list -- the client is unable to provide the service. A failure on part the of the IETF to define an adequate mechanism to address this problem has a very high probability of causing individual implementors to develop their own solution on an implementation-by-implementation basis. Even if a sufficently large critical mass of implementations begin using the same convention, there will almost certainly be a substantial period of time before a widespread pattern is established. Until such a de-facto standard is established, interoperability between independant implementations will suffer. Further, even if the convention for such a mechanism is eventually established, older, non-interoperable conventions will continue to exist side-by-side with it indefinitely for reasons of backwards compatibility.

Currently, the accepted solution to this problem is that such information is manally entered into the client by the user. While this invokes only a mild startup cost whenever Bob goes to set up his home PC (not entirely unlike configuring the POP and SMTP servers for an e-mail client), it adds an extra step to Bob's login process when he's in an internet cafe, at a friends house, or at any other device that he doesn't use on a regular basis. Chances are very good that Bob isn't going to want to remember the additional URI for his buddy list -- or, even if he can, probably doesn't want to go through the trouble of typing it in (in addition to his user ID) every time he logs in from a different location. Requiring him to do so provides an experience that is clearly inferior to those available from proprietary solutions today. In short, while the approach of requiring the user to enter an additional URI to access his buddy list is a solution to the problem of where the information comes from, it does not do so in a way that is, from a user perspective, as good as currently available products. Because of this added inconvenience, implementors will likely attempt to solve the problem in a variety of non-interoperable ways, as discussed above.

One approach to solving this problem is to establish a convention that indicates how to manipulate the URI in such a way that it indicates the resource to which the SUBSCRIBE should be sent; for example, appending "-buddies" to the user portion ("sip:bob-buddies@example.com") would be one such transformation, as would using the hostname portion (e.g. "sip:bob@buddies.example.com"). While acceptable from a technical perspective, this approach runs afoul of several philosophical objections and has some suboptimal characteristics. The prime philosophical objection is the supposed property that URIs are (with certain well-defined exceptions) treated as opaque by clients who use them. Establishing a convention that describes specific transformations of the URI violates this property. Suboptimal characteristics of any implicit approach include relics such as requiring the user's registrar to handle buddy list services and limiting users to having a single, centrally managed buddy list.

Another approach to solving the problem under discussion is to allow the URI for the buddy list itself to be retrieved from the user's home domain server (e.g. example.com). Doing so provides an explicit way of indicating from where to retrieve the list. This approach is, in spirit, similar to that defined for device configuration ; specifically, a subscription would be sent to the user's address-of-record for an event package that contains user configuration data. One component of the user's configuration information would be a URI (or possibly even URIs) that indicate from where the user's buddy list could be retrieved. In addition to providing a clear mechanism for unambiguously identifying a user's buddy list, this mechanism has the additional properties that it allows buddy lists to be hosted by a domain other than that of the user's registrar, and that it allows users to have multiple buddy lists configured. Finally, this approach can be specified in such a way that it allows inclusion of additional user-profile information if needed, such as a URI for message waiting indication .

Thanks to Paul Tidwell for first raising the issue discussed in this document. Steve Donovan, Robert Sparks, and Dean Willis contributed to early conversations on the topic.