By James Bagley
[Deni’s note: I asked Jim Bagley, senior analyst with Storage Strategies NOW to write this blog on parellelization of applications.]
During one week in late July, a team at Purdue University built one of the world’s most powerful supercomputers between breakfast and lunch. The system included 1,200 HP dual quad-core computing nodes connected with 10Gigabit Ethernet fabric. The resulting machine was named the Coates Cluster after Dr. Ben Coates, who headed the Purdue School of Electrical Engineering for 10 years beginning in 1973. Coates was instrumental in developing Purdue as a leading computer technology facility, and the machine is estimated to be in the top 50 of High Performance Computers worldwide.
The same week we had the opportunity to discuss the challenges of optimization of applications within these multi-core systems with Dale Geldart, COO of Exludus, a company that is focused on the development of software architecture to automatically address the issue.
“Most applications work serially, which is not the most efficient way to access the capabilities for parallel processing within multi-core systems,” Geldart says. “And, parallel programming techniques are too complex for most development environments.”
Exludus, based in Montreal, has taken the approach of building a middleware layer between the application and the operating system, which analyzes the resource utilization and stages functionality to maximize parallel operations. In a multitasking environment, applications are automatically forced to collaborate on resource requirements, creating a dramatic increase in throughput in all applications. By taking advantage of the knowledge of one application waiting for a needed resource to let another complete the task and free the resource, the multi-core optimizer is like a supercharger for supercomputers.
One of my favorite professors at the University of Utah was Dr. James Case, who pioneered in the science of parallel computing, and envisioned machines like the Coates Cluster. He even developed programming techniques to take advantage of such systems. That was over 40 years ago in the late 1960s. Dr. Case died unexpectedly at a relatively young age in 1990, just when advances in neural network hardware development were beginning to take advantage of his well-developed processes and research into artificial intelligence.
We were able to track down the granddaughter of Dr. Case, Amber I. Case, who is specializing in the field of cyborg anthropology. The field specifically analyzes interaction between humans and machines. Amber is the daughter of Elliot Case, an accomplished scientist and inventor in the field of acoustics, with upwards of 50 patents.
“I grew up thinking that everyone had a laboratory in their basement,” says Amber, bemused. Already an expert mathematician and electronics technologist by high school, Amber studied Sociology and Anthropology at Lewis and Clark University. “I could make machines do anything, but had trouble communicating with people, so I wanted to work in the social sciences for my undergraduate degree.”
She remembers Dr. Case as an incredibly fun person, who would build exotic and beautifully engineered kites and loved to spend weekends with his family. She also was, as a prodigy, amazed by his works dating to the 1960s.
“I remember reading his papers on three-dimensional processing when I was in high school, and thinking that he was at least 50 years ahead of his time.”
We are confident that Amber, who wants to continue her post-graduate work at MIT, will carry on the family’s tradition of technology advancement.
