Meeting Emerging Challenges in Computer Architecture
In 2003 Professor Per Stenström, Department of Computer Engineering, Chalmers University of Technology started a STINT supported cooperation with Professor Michel Dubois, Electrical Engineering Systems, University of Southern California, USA within ‘Computer Architecture’.
Computer architecture is the engineering science of computer design. It is mainly concerned with principles for how to transform the raw power of the technology into computational speed. While technology improvements have yielded an annual performance improvement of 35%, as forecasted by Gordon Moore in the late 1960s, advances in computer architecture have led to an additional performance improvement of about 30% annually. To bring this into perspective, as many as 10 billion instructions per second are computed in today’s computers as opposed to a thousand in the first electronic computer in the 1940s. Moreover, while the first electronic computer occupied a room of approximately 100 square meters, today´s computers are squeezed in on a chip consuming an area of 1 square centimeter. Consequently, performance and size have both improved more than six orders of magnitude.
There are however three main challenges for how to build faster computers: (1) how we can exploit more parallelism in computer programs (2) how we can supply computers with instructions and data faster, and (3) how we can evaluate the effectiveness of new principles. In terms of finding more parallelism, it is essential to consider longer sequences of instructions so as to identify a larger number of instructions that are independent and can be executed in parallel. In order to supply these instructions and their data faster, new principles for managing fast and small memory buffers – called cache memory – becomes essential. Finally, innovative design principles to make this happen must be evaluated using accurate models. Another hot topic in computer architecture is therefore modeling techniques.
STINT has financed a collaborative research and education program in computer architecture between Chalmers University of Technology and the University of Southern California since 2001. The two research groups have collaborated informally in the past. STINT has made it possible to strengthen the research and education programs at the two sites through international student and faculty exchanges. Since its start, six Ph. D. and Master´s student exchanges have taken place resulting in many joint research activities. Additionally, education experiences have been shared at the faculty member level resulting in a boost in the quality of the offered courses in computer architecture at the two institutions.
Research has targeted the three challenges outlined above. The joint research has contributed with design principles to make it possible to uncover more parallelism by more accurately predicting which instructions will be executed in the future. The research has also contributed to principles for cache management. The discovered principles have made it possible to use the cache memory more effectively resulting in fewer accesses to slow memory. Finally, in another joint project, a simulation system has been developed that makes it possible to study the effectiveness of innovative design principles in a significantly more realistic context than has been possible before. This effort has gained a lot of attention in the research community. The exchange program has also contributed to the curriculum in computer architecture. Apart from exchanging experiences and course development efforts, the faculty members leading this effort are now collaborating on a forthcoming textbook in computer architecture.
Senast uppdaterad: 04-10-12 13:17