Apple World Wide Developer Conference - 2007

Often a site of industry-changing events, Apple's Developer Conference brings hardware and software developers and vendors together to learn the latest plans from Apple for the Macintosh platform and exposes its attendees to the latest technology from Apple and from other developers.

To achieve accessible computational power for our research goals, we developed, on the Macintosh platform, the tools to build easy-to-use numerically-intensive parallel computing clusters. We find that the usability and reliability of the Mac cluster technology is as important as its performance. Our approach is designed to allow the user, without expertise in the operating system, to most efficiently develop and run parallel code, enabling the most effective advancement of scientific research. By "reinventing" the cluster computer, we provide a unique solution designed to maximize accessibility for users.

To support numerically-intensive and tightly-coupled problems that demand the computational power and networking capabilities of clusters of Macintoshes, our software technology supports five implementations of the Message-Passing Interface (MPI), today a dominant industry standard. MPI's status implies how its underlying computing paradigm has revealed itself to be the most efficient and economical way yet found to apply large numbers of processing "cores" effectively for general purposes.

We present two new applications of our approach to clustering:

1. The new Supercomputing Engine for Mathematica enables Wolfram Research's Mathematica to be combined with the programming paradigm of today's supercomputers. In contrast to typical master-slave "grid"s, this solution instead closely follows MPI, from inside the Mathematica environment, and has every kernel in the cluster communicate with each other directly and collectively, necessary to address the largest problems in scientific computing.

2. Modern compression/decompression (codec) algorithms for video processing increasingly have compression times that vastly exceed playback times. We address this emerging computational demand using MPI-based cluster computing. We implement load-balancing parallelization of QuickTime video compression, including frame-reordering H.264, and interface our cluster support software with mainstream desktop video-editing applications such as Final Cut Pro.