Xian-He Sun, Professor of Computer Science at Illinois Institute of Technology and creator of Sun-Ni’s law—one of three scalable computing laws along with Amdahl’s law and Gustafson’s law—has established a new mathematical model for reducing data access delay.
Called 'Concurrent Average Memory Access Time (C-AMAT)', it could cut the penalty associated with accessing data and increase speed by up to 100 times through parallel memory access, which in turn will create a ‘break’ in the memory-wall problem.
During the last four decades, CPU speed increase has been following Moore’s law, increasing 52 per cent per year and doubling every 18 months. However the speed of memory is only increasing nine per cent per year, and disk speed is even further behind, increasing an average of only six per cent per year. Memory speed currently is about 400 times slower than CPU speed. This is what is known as the memory wall for data movement and processing.
Sun’s recent work on C-AMAT is the first formal mathematical model to promote and evaluate the concept of parallel memory for reducing data access delay via explicit parallel data access.
C-AMAT is a vital tool to mitigate the memory-wall effect and to improve memory system performance. AMAT states that if the desired data is in cache (hit), then you get the data quickly; otherwise (miss), you get a cache miss penalty. Due to the memory-wall problem, the miss penalty will be big. So architecture and algorithm design focuses on reducing cache misses. With C-AMAT and parallel memory access, however, depending on if there is a hit occurring at the same time, a miss may or may not have a penalty. C-AMAT would change the focus of architecture and algorithm design from reducing cache misses to increasing data access parallelism. It provides a formulation to evaluate the effectiveness of the concurrency of each memory layer toward the final performance of parallel data access.
‘The most profound research is not the design of the fastest algorithm for a given problem; it is revealing a fundamental computing property so hundreds or even thousands of algorithms can be developed upon it,’ Sun said.
Sun has been working on memory access issues for 20 years. During this period, his research has been continuously supported by the National Science Foundation (NSF) and other government agencies. A paper on Sun's results will be published in the in the Institute of Electrical and Electronics Engineers (IEEE) Computer Society's magazine in the near future.