UDDI Business Registry

» see uddi.org as it lived, on the wayback machine

~ the goodbye ~

I met this as a diagram on a lecture slide. Turkish computer engineering syllabi picked up web services in the middle of the 2000s and the drawing was always the same, a triangle with a registry sitting at the top corner and two companies at the bottom finding each other through it. Nobody in that room, including the person at the board, had ever actually queried the thing. By then it had already stopped accepting new entries.

UDDI was announced in September 2000 by Microsoft and IBM, with Ariba alongside them. The specification described how to write down a business and its services in a form another machine could read, and the Business Registry was that idea running in public as a reference implementation. It went live on 2 May 2001 after a beta period, with IBM and Microsoft each operating a node and Hewlett-Packard signing on to run a third.

The nodes replicated to one another, so publishing into one put you into all of them. That replication was the point being proved. Several independent implementations of one specification, operated by companies who competed on everything else, holding a single consistent dataset between them. There were over fifty thousand entries by the end and the interoperability worked exactly as designed.

What did not work was anybody wanting the entries.

~ the chronicle ~

~ what it was supposed to do ~

The vision was automatic business relationships. Your purchasing system would search the registry for a supplier that spoke the right protocol and read off the technical details of how to connect to it, then begin trading without a human having negotiated anything first. In 2001 that was written down seriously in a great many places.

The failure was not technical and this is the part worth keeping. Companies do not choose suppliers by lookup. They choose them through contract negotiation and through somebody flying out to see the factory, and the connection details are the last five percent of a process that is mostly legal. UDDI automated the five percent and assumed the rest had already been handled.

~ why it did not work ~

There was no gatekeeping because there was no money. Nobody paid to run the registry, so nobody was funded to check what went into it, and anyone at all could publish. The result was a directory full of test entries and abandoned records, which is the state every free unmoderated directory eventually reaches. Microsoft's own engineers described it in those terms when the shutdown went out.

The announcement was posted on 16 December 2005 and publishing stopped on 12 January 2006. The wording was careful and it was fair: the interoperability had been demonstrated and the specification was ratified at OASIS. Both of those are true. It is also true that a public directory closing because nobody searched it is a different event from a proof reaching its conclusion.

UDDI itself did not die that day. Microsoft shipped UDDI Services inside Windows Server 2003 and IBM put a version 3 registry into WebSphere, and companies ran private registries behind their own firewalls for years afterward, where the governance problem solves itself because one organization owns every line in the list. The global version is the part buried here. What replaced it, in the end, was a documentation page and an email address.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »