Research for New Technology Using Supercomputers

Fujitsu has been vigorously working on the development of supercomputers since the 1980s, mainly focusing on large scale integrations (LSIs), systems, software including operating systems and middleware and mathematical libraries. Meanwhile, the development of applications such as computer simulations has been left to external research institutions including universities and national laboratories and independent software vendors (ISVs). However, for effective utilization of massive large-scale parallel computers such as the K computer,note)i a certain paradigm shift is required in application development as well. That is, as known by Amdahl’s law, the ratio of the sequential fraction has a major impact on the scalability of a computer when a program is

run in a parallel environment. As a result, the program itself ends up being unable to effectively use computation nodes even if they exist in abundance. The K computer has 640 000 cores and, to achieve a scalability of about 90%, the execution time for the sequential fraction must be reduced to about 1/5.8 millionths of the whole or less. Except for simulations using Monte Carlo methods or such like, the development of such a parallel program requires us to create new computation techniques based on thorough consideration of computer architecture. The key to doing that is to combine computer development technology and application technology.

Accordingly, since its initial development of the K computer, Fujitsu has embarked on research and development of applications as it aims to solve these challenges and achieve effective utilization, in a true sense, of super- large-scale computation systems like the K computer. For improved efficiency in these activities, Fujitsu has cooperated whenever possible with universities, enterprises and national research institutions already engaged in leading-edge research to move forward with the development of a super-large-scale parallel program.