Tuesday, February 12, 2008

A Bulding Block Approach to Standardization

For decades, the telecommunications industry has standardized solutions from A to Z, with little if any reuse of existing specifications when creating new ones. The progressive migration from circuit switched to IP based services did not initially change this fact much: MMS or OMA IMPS (that I take as an example in this post) are typical examples of creating telco-specific standards based on a loose reuse of IETF ones (SMTP for MMS, HTTP for OMA IMPS).

This has changed with IMS, and more especially its SIP component. 3GPP and the IETF collaborate with each other, and needed extensions to the SIP protocol due to IMS requirements are under the control of the IETF.

By importing IETF specifications into telecom standards, 3GPP implicitly accepted the building block approach to specifications that is common place in the Internet domain. In this post I will try to describe this approach and its benefits.

Building block standardization of SIP

SIP is a textbook example of a building block approach to standardization. The people and groups in charge of specifying SIP constantly try to apply the following rules:
- Do not reinvent the wheel. Reuse and adapt existing specifications if they fulfill your requirements. Only create when needed.

- Make everything as generic as possible. Even if your requirements are very precise, try to make your solution generic enough to be reused for other requirements.

Here follow some examples of how this was applied to SIP standardization:

- SIP sessions make use of the Session Description Protocol (SDP), which was specified prior to SIP. In effect, it is possible to use SDP without SIP.

- SIP SUBSCRIBE and NOTIFY methods were initially created to support a very specific requirement, actually related to the telecom domain (the support of the telephony Automatic Call Back service with SIP). However, it was decided to make the concept a generic and extensible means to distribute event notifications in a SIP network through event packages (see the first draft for SUBSCRIBE/NOTIFY here). When a part of the IETF community decided to support presence through SIP, they simply had to reuse the event package specification and create two presence-specific event packages. While the requirement was initially very specific, it gave birth to a concept that is fundamental for SIP and constantly evolving through the creation of new event packages. It is actually remarkable that this is a telephony -related requirement that led to a SIP concept which opens the door to a large variety of non-telephony related applications of the protocol.

- In the Instant Messaging (IM) area, presence was initially no more than a single state, describing if a recipient could accept an IM. The IETF decision to support presence through the inclusion of an XML document in the body of SIP methods, and allowing extensions to the basic schema, permitted the definition of presence to be gradually extended to become a large set of information about users (or services), their communication means, terminals and applications.

- SIP PUBLISH was initially created specifically for a client to remotely update presence information. The first versions of the draft were tightly linked to the presence event package and made impossible the reuse of PUBLISH in different contexts (see the very first draft here). However, the IETF community rapidly ensured the possibility to reuse PUBLISH for all existing and future event packages. PUBLISH therefore contributed to the enrichment of SIP-based presence, but at the same time a requirement initially scoped to presence contributed to the enrichment of the whole SIP protocol.

- Instant Messaging through SIP was initially supported only through the creation of a new SIP method: MESSAGE. However, it rapidly emerged that this approach was far from optimal to support all potential requirements associated to instant messaging: the concept of chat, which embeds IMs in a specific dialog context, the need to potentially exchange large documents via IM (e.g. a video file) while SIP is a control protocol and not a transport one like HTTP, or the need to support potentially high IM traffic while a SIP infrastructure might not have been implemented with this purpose in mind. It took time and several tries for the IETF community to address these requirements, and the final decision was to reuse the concept of SIP session as well as another protocol to transport an IM within the session. As a protocol like HTTP was not optimal to support the requirements for this IM transport protocol, it was decided to specify a new one called MSRP. This decision makes the comparison between Jabber/XMPP and SIP to support IM very biased. Maybe Jabber/XMPP is a better protocol than SIP for IM. However, Jabber/XMPP was initially specified and optimized for it, making its extension for, say VoIP, far from straightforward. On the other hand, in a SIP context, IM can be perceived as one communication component among others in a multimedia session.

OMA IMPS vs. IETF Presence and IM

The vertical standardization mindset that still prevailed a few years ago in the telecom community can be illustrated with OMA IMPS (initially called Wireless Village), a mobile specification to support instant messaging, chat rooms and presence.

Instead of reusing IM and presence related protocols available in the Internet, the Wireless Village group decided to specify a client to server protocol and a server to server protocol that would be specific to the mobile telecom domain, just reusing HTTP as a semantic-less transport protocol for OMA IMPS commands.

The group also decided to define IM, chat rooms and presence as tightly coupled together from a protocol and an architecture perspective, and to tightly link presence information to the mobile context.

In order to support its requirements, the Wireless Village group had to define various kinds of user lists (or groups) serving different purposes. Instead of creating a generic user group concept, they decided that each group fulfilling a specific purpose was a distinct object. Consequently, each group object led to a set of specific commands in the protocol, for creating/deleting the group, adding/removing elements to it, etc. With such an approach, if you define, say 4 types of user groups and 6 management commands, you end up with 24 distinct commands in the protocol.

In comparison, to address similar objectives, the IETF decided to decouple various concerns.
While presence is a concept originated in an IM context, the IETF decoupled one from the other, permitting each to evolve independently, thus permitting presence to apply to a much broader scope than simply IM.

By reusing the SIP session concept for session-based IM, the IETF permitted both the implementation of IM-specific systems, and multimedia systems using IM as one component among others in a SIP session.

The approach to address user groups and associated management, specified in RFCs related to XCAP, followed this approach:
- A user group is a user group, no matter what it is used for. The same user group can serve different purposes, and the set of applications for user groups is not arbitrarily bounded.
- A user group is user data, and there might be other user data that require similar access and management. No need to specialize access and management methods to user groups.
Consequently, XCAP is an HTTP-based protocol defining a few data management methods. The data itself is specified in XML, and there exist specifications for these data being user groups. As one of the requirements associated to data management was to be able to notify a user about changes made to data, the IETF decided to use a SIP event package. In effect, the IETF specifications for user data management include the joint usage of XCAP and SIP.

Building block standardization approach in IMS

The building block mindset to specifications has spread to IMS and non IMS standardization into 3GPP.

For instance, despite a terminology which is heavily related to SIP sessions (e.g. CSCF - Call Session Control Function), the IMS core network can be seen as a SIP connectivity network able to route SIP signaling, whether it is session-related or not, within an IMS domain, across IMS domains, and between IMS and non-IMS SIP domains.

In this context, IMS Presence, Messaging, and Chat Rooms are implemented as independent applications on top of the IMS core network and that make use of it. Once again, the comparison with OMA IMPS is quite interesting:
- OMA IMPS specifications lead to an implementation based on a network of IMPS servers over the mobile IP network. An IMS implementation relies on deploying application servers on top of an IMS core network. The IMS core network directly supports some of the requirements that are supported vertically in the OMA IMPS specifications (and implementations), like user authentication or routing and interfacing between various operators' OMA IMPS networks.
- OMA IMPS specifications tightly link the concepts of IM, presence and user groups. On the other hand, IMS specifications treat each of them as independent enablers which can be used together or in different contexts.
- OMA IMPS specifications were totally under the control of the Wireless Village group, and then OMA. On the other hand, by reusing IETF specifications, IMS specifications directly benefit from the evolutions performed in the IETF community, including some originating from people and companies which do not belong in the telecom or IMS domains.

A quite similar comparison can be applied to MMS and the equivalent support through IMS messaging.

Another interesting example is the 3GPP Generic User Profile specification, which permits to provide a centralized and homogeneous access to user data actually residing in various locations (e.g. HLR, HSS, AuC, application servers) and normally accessed through a variety of protocols (e.g. MAP, Diameter, LDAP). At the beginning of the erratic standardization process for GUP, 3GPP intended to standrdize a specific GUP protocol as well as a specific GUP schema to describe user data. Later on, it was decided to align on the specifications for Liberty Alliance, which define web services permitting 3rd party service providers to access user data owned by the network operator. As a consequence, GUP can be used directly as the means to access user data in network databases to support the Liberty Alliance web services exposed to 3rd parties.

The GUP specifications were also made generic enough to clearly distinguish between the methods used to access and manage user data and the data itself, that needs to be specified by instantiating and extending the generic GUP schema. On the other hand, the GUP specifications also include a SOAP-based user data modification notification mechanism, which duplicates what SIP event packages can and do support for XCAP. However, one can argue that the usage scope of GUP is broader than IMS and cannot rely on a protocol that 3GPP only uses in the context of IMS.

Some advantages associated to building block standardization

Reusing existing specifications instead of defining them from scratch permits to speed up the standardization process.

A protocol component or an application performing a generic task can be implemented once and reused several times, leading to faster development and validation.

In some cases, building blocks can be re-arranged with others to create new solutions. I gave the example of session-based messaging which, by applying the concept of SIP session to instant messaging, permits to integrate IM as one component among others in a multimedia SIP session.

Christophe

4 comments:

Ran Arad said...

While the building-blocks approach is great, the case for SIP is not so good: the foundation needs to be solid enough to allow building on, and I do not think this is the case.

SIP was initially meant to to be a simple way to initiate an audio session. The things that are expected from the protocol now make it far more complex than it is able to be, and any minor interoperability issue is magnified with every block you add.

I recently blogged about complex text-based protocols, see .

Christophe Gourraud said...

I think that flexibility, extensibility and power usually come with new issues and challenges, that you are pointing at concerning SIP.

It is usually easier to comprehend and control monolithic concepts, while distribution and modularity come with new potential but also with a new complexity.

You can apply this to SIP, but also to the whole IMS service architecture, as well as the most important advances in the IT domain in the last 20 years.

Christophe

Debra said...

The dude is completely just, and there is no suspicion.

Anonymous said...

Hi! I know this is somewhat off topic but I was wondering which blog platform are you using for this website?
I'm getting tired of Wordpress because I've had problems with hackers and I'm looking at alternatives for another platform. I would be great if you could point me in the direction of a good platform.

Feel free to surf to my weblog :: waist to hip ratio