We have all seen the overlay graph of chip transistor density and the growth rate of GenBank, where the GenBank curve rapidly accelerates past the transistor density curve. One may deduce that the power required to process genomic data is not keeping pace with data production. PCs are fast and getting faster, but will not likely surpass scientific computational demands with present silicon chip technology. For the most part, high-throughput computational biology processing takes place in the data center. From a computational biologist's perspective, we are still enslaved in the highly complex data center model, and we pretend to be happy about it.

A PCC is a loosely coupled, distributed computing platform that fits in or near a personal workspace. It has no special power requirements. Any supporting infrastructure, such as switches, UPS, and cabling, are preconfigured. The overall hardware is self contained and reasonably quiet.

The PCC must be preconfigured to enable the scientist the moment he or she pushes the "on" button. Just as the DOS prompt proved insufficient, the PCC must be an out-of-the-box scientific platform already loaded with base cluster-enabled applications. The platform must be secure, extensible, and able to integrate with other data/network resources.

1977	2004
Taking control: Like the revered Apple II, personal compute clusters such as the Apple Workgroup Cluster enable researchers to compute outside the total control of the data center.

PCCs must be priced so they fit within discretionary budgets of up-and-coming faculty or research staff who may be building a research group from scratch, such as Edward DeLong at MIT. PCCs are also used in bioinformatics teaching environments, such as Mike Thomas' lab at Idaho State University, to provide hands-on access to modern compute resources. In most cases, the target end-users are scientists, not IT administrators. Therefore, many of the administrative tasks involved in compute clusters must be encapsulated and simplified for non-IT professionals.

Several hardware vendors have recently become sensitive to this subtle PCC market pressure. Apple Computer again is the user-empowering trailblazer with its Apple Workgroup Cluster for Bioinformatics. This G5-Xserve-based cluster has succeeded in the category of ease-of-use and optimal design. This past February, Sun announced the release of its Sun Fire Starter Cluster, perhaps the most powerful and economical PCC solution currently available.

The most recent innovative PCC solution comes from a new company called Orion Multisystems. Orion has just announced the availability of its desktop cluster workstation — a 12-node integrated cluster architecture encapsulated in a single case slightly smaller than a common tower box. The machine uses less than 200 watts of power under maximum load from a standard power outlet and can be extended up to 48 nodes (or up to 96 nodes with the deskside system).

As an IT administrator, your scoff at the idea of a PCC may echo your predecessor's PC scoff uttered 25 years ago. Empowering computational scientists with (somewhat) scalable computing resources that supersede the power and economy of the most powerful workstations may be the next logical step. The creative scientific process sometimes runs a bit faster than the "on-demand" resource requisition bureaucracy in many large organizations. Scientists crave control of their own computational resources — perhaps toward distraction.

Michael Athanas is a founder of The BioTeam, which delivers scalable informatics solutions in collaboration with its partners, Apple, Sun, and Orion Multisystems. E-mail: michael@bioteam.net.