This case involves an intellectual property dispute relating to high-performance computer (HPC) systems. It was alleged that a relatively small company in the space infringed on several of the primary patents which address the hardware and software aspects of HPC systems. An expert with an understanding of HPC system hardware and software was sought to discuss the different configurations of CPUs and switches as well as fault protocols within an HPC system and java virtual machine (JVM) performance.
Question(s) For Expert Witness
- 1. Please describe your experience researching or working in the field of high-performance computing systems.
Expert Witness Response E-066468
I have been working with HPC systems for over 12 years. During that time I have covered all roles including hardware selection, procurement, and installation (system design). I have also dealt with user application support for parallel programming and system tuning. I have held leadership roles setting HPC strategy and next-generation design/needs for the institution. I am very familiar with job scheduling software as a key component that makes a cluster a cluster. Over the years, we have used OpenPBS, PBSPro, Maui/Moab, and evaluated others such as SLURM and HT-Condor. The scheduling stack is key for implementing our policies and getting our expected workload. HPC systems commonly use two networks, a management network (normally classic Ethernet) and a data and communication network for MPI (message passing) and filesystem (lustre/GPFS/PVFS). I am most commonly familiar with Mellonox Infiniband, but have also used Miyricom and and Qlogic (now Intel Omnipath) interconnects. I have also read extensively about the CRAY interconnects. I have lectured to graduate students and researchers how the clusters that we have supported work.