April 30, 2008
Solving the massively-parallel software problem
Intel and Cray have been talking this week about building supercomputers with a million cores or brains, but how will all those processors-within-processors work together and communicate with one another and how difficult will it be to write applications that take advantage of all of them?
This is the question that Stanford University hopes to answer with its Pervasive Parallelism Lab, announced on Wednesday.
“Parallel programming is perhaps the largest problem in computer science today,” said Bill Dally, chair of the Computer Science Department.
“[It] is the major obstacle to the continued scaling of computing performance that has fuelled the computing industry, and several related industries, for the last 40 years.”
The problem has arisen because multi-core processors were too expensive until recently for all but high-performance computers, meaning few programmers have developed the expertise to take advantage of their new, affordable abundance.
Stanford says computer scientists fear the progress of computing could stall and that’s clearly a worry for the major microprocessor suppliers - Intel, AMD, IBM, Nvidia, HP and Sun are all supporting the new lab, which will pool the efforts of the university’s leading computer scientists.
It is also not the first initiative of its kind. Intel and Microsoft announced last month they were investing $20m in research at the University of California at Berkeley and the University of Illinois at Urbana-Champaign in the same area.
Stanford’s lab has a budget of $6m over three years. The aim is to develop a complete parallel computing system, covering everything from fundamental hardware to new user-friendly programming languages. This could lead to the system doing all the work for developers to optimise their code for parallel processing.
