Thursday, October 1, 1998

Massively distributed computing using computing fabrics


By David Gewirtz

There's a funky, little think tank in Sedona, Arizona that thinks it has identified the next wave of computing, what it calls "computing fabrics". Fundamentally, the computing fabrics concept is another take on the idea of massively parallel systems, but in this case these systems can be reconfigured dynamically. The key thinkers behind this new architecture definition are Erick and Linda Von Schweber of the whimsically-named Infomaniacs.

As Notes and Domino administrators, you're viscerally familiar with the concepts of distributed computing, at least in the non-transactional sense. With a distributed database architecture and replication at the very core of Notes, the Notes/Domino pair is arguably the most successful implementation of a distributed database environment.

The difference between Notes' replication and computing fabrics' form of distributing computing is, in part, the degree of couplehood. Two key concepts from the world of multi-processing are "loosely coupled" and "tightly coupled". I'll discuss these terms in more detail throughout the article.

Notes (and, in fact, the Web) are loosely coupled processing environments. These systems easily share information, as if they're part of one large body. That's why we talk about the Web as "THE Web", not "all them webs". They're connected, but they're not sharing computing or memory resources.

What is a "computing fabric"?

Although the definitions can get much more complex, a tightly coupled multi-processing environment is often one where the memory address space is shared among, and addressable by, the processors, in addition to various other computing resources.

The idea behind computing fabrics is that processors participating in a computing fabric are somewhat fluidly-coupled. Fabrics share many of the attributes of a loosely coupled network, including the ability to be disconnected and reconnected dynamically. But fabrics also create an environment where the address space across machines appears to be homogenous. This is where things can begin to get interesting.

The Von Schwebers define computing fabrics thusly: "Computing Fabrics are richly interconnected processors, memory, and storage out of which tightly coupled systems, that then loosely couple, can dynamically form, uniform, and reform."