Leibson's Law

A lot of what you read about multicore these days is heavily theoretical and light on practical details. Frankly, there’s a lot of unsubstantiated opinion and downright hype out there. Refreshingly, this practical article in NASA Tech Briefs (How a Supercomputing Innovation Accelerated a Business) describes how a cluster of 64 Intel Core-2 Quad processors linked with Gigabit Ethernet connections has accelerated 3D CAD for a quick-turn plastic parts company called Proto Labs. The article describes how the cluster’s 256 processors generate 3D tool paths for the company’s milling machines so that prototype plastic parts can be fabricated in a day. Proto Labs built the cluster because the success of their First Cut Prototype division was stretching their existing computers and turnaround time was becoming a problem.

Here’s a quote from the article that explains the situation: “A couple of years ago, we recognized an exponential increase in the demands on our computing resources. This was attributable to growth in demand for our services due to our geographic expansion, as well as the enhancements we made to our processes to support bigger and more complicated part geometries. It became apparent to us that before long we would hit the proverbial wall and we needed to take action.”

And here’s a description of the result: “Less than three years ago, this cluster would have been one of the world’s 500 fastest computer systems. Less than 10 years ago, it would have been the fastest in the world. Today, it is merely a very powerful supercomputer managing internal and external operations at levels that would be virtually impossible without the high-performance cluster.

The end result is that we smashed through the wall, and despite the continued ramp-up in quoting and manufacturing demands, we were able to make our quotations and deliveries even faster...”

It is possible to solve bigger problems with multiple processors. The key is to understand your problem and then to apply the multiple processors appropriately. What’s true today for Proto Labs working with a rack-sized computer cluster was once true for room-sized supercomputer installations and will soon be true for single-chip SOCs. The scale of the processors may change but not necessarily the architectures. (Don't miss Paul Rako's entertaining blog on the topic.)