John R. Baumgardner, Ph.D.                                                                                                                                            September 1, 2008

Ramona, CA 92065

To whom it may concern:

This is a letter of recommendation for Arch C. Davis of Princeton, New Jersey. In regard to my own background, I have a Ph.D. in geophysics and space physics from UCLA and spent 20 years in the Fluid Dynamics Group of the Theoretical Division at Los Alamos National Laboratory engaged primarily in large-scale numerical modeling, mostly on large parallel computer systems. I retired from Los Alamos in 2004 and until recently have been employed by a private research institute, also engaged in large-scale scientific computing.

I have known Arch since 1968 when together we entered Princeton’s Department of Electrical Engineering as first-year grad students. We have kept in contact as friends since then, but about ten years ago we began interacting on the topic of high-reliability computer cluster design. This led me to introduce Arch to a former student of mine, Dr. Peter Bunge, who in 1998 had accepted a faculty position in Princeton’s Department of Geology and Geophysics. Peter’s primary research interest was modeling the dynamics of the earth’s mantle in full 3D spherical geometry, which requires significant computational resources. Peter’s interaction with Arch led to Peter’s decision to work with him to procure a large custom-built cluster. This cluster, engineered and built by Arch under contract with Princeton in 1999 for high reliability, consisted of 140 Intel Pentium II processors and was named Geowulf. Its exceptional reliability is attested by the fact that it ran for 6 ½ years with almost no component failures. Arch designed and built Peter a second custom cluster in 2002, named Bladerunner, consisting of 128 much more powerful Intel Pentium IV processors. It likewise proved to be a workhorse machine and is still in operation.

Having become keenly aware that such custom clusters were highly cost-competitive relative to systems then being marketed by the major computer vendors, in 2002 I purchased such a cluster from Arch for my computational fluid dynamics group at Los Alamos, primarily for running climate simulations. This cluster, named Epiphany, consisted of 72 Intel Xeon processors configured very compactly in 36 dual-processor blades, designed by Arch, and interconnected through two 24-port gigabit Ethernet switches. Its total cost was only $125,000. Arch’s use of high-quality ball-bearing fans, together with high-quality disk drives and power supplies, resulted in a system that likewise has experienced almost no hardware failures so far during its lifetime.

With these positive experiences in view, when I joined a private research institute after retiring from Los Alamos, I proposed the procurement of a similar cluster to meet the computing needs of the institute. Again we contracted with Arch, this time for dual-processor blades using even more powerful AMD Opteron processors. This system, delivered in mid-2005, consisting of 44 total processors cost only $65,000. It also has proved extremely reliable and almost maintenance free.

These experiences have given me a high regard for Arch’s exceptional technical competence and also an unusual attention to detail that is so critical in the engineering of computer systems that are reliable, meet the performance specifications, and deliver true customer satisfaction. I also have been deeply impressed with Arch’s capacity to grasp the bigger picture to identify fundamental ways that computer designs can be improved, ways that others seem to overlook. I therefore have no reservation in giving Arch an extremely favorable recommendation as lead designer and system architect in a venture to develop and bring into production energy-efficient computing technologies.

John Baumgardner