There’s been a long-standing competition between centralization and distribution, two important but opposing ways of designing computer and communications systems. With cloud computing (hosted services in mammoth data centers) the pendulum is moving back to centralization. There’s also been a similar swinging of paradigms in the telecom world.
Who doesn’t have the urge to centralize control? It’s just a natural way to manage everything from hub-and-spoke communication networks, world-class kitchens (think Gordon Ramsay, Mario Batelli) computer (client-serverr architectures), to transportation routes.
Our gut instincts favoring a single (or just a few) points of management also have some practical payoffs. Maintenance issues are easier—no need to update hundreds of subordinate process. And with centralized data, consistency gotchas are eliminated or greatly reduced . On the other hand, distribution of resources improves performance, scales better with increasing demand, and leads to better allocation of resources.
In the Internet era, there’s room for both ways. It’s a question of degree. A great example of a hybrid approach is the Distributed Name Service, aka DNS, the servers that resolve “www” names to more practical IP addresses. DNS is made up of centralized data bases that are assisted by second-level of distributed servers that are in private hands.
Back in the late 1980s and early 1990s, our public phone network was undergoing its own centralization episode. With the implementation of the “intelligent network“– an architecture developed by Bellcore–centralized computer control was added to what was at the time a distributed phone network.
The intelligence came in the form of mainframe computers that could dip into an database to translate phone numbers. It’s a simple concept–a “service control function” in IN terminology– but leads to powerful applications like toll-free 800 numbers, local presence, and calling cards. (Editor’s note: Hey, the centralized IN never reached anything like the level of application flexibility of the more loosely organized Internet.)
Taking some ideas from the Intelligent Network, IMS , for IP Multimedia Subsystem, may be the next evolutionary step of the IN, adding even more centralized application control to handle video, speech recognition, billing instant messaging, ….
Networking sage Brough Turner is for me the last word on whether IMS ‘s centralization will make for a voice-video-data network with lots of applications, like, you know the Internet. Turner makes the point that one of the prices of centralization is complexity—you have to really understand what’s going in the core to get your application up and running.