|
Hiroshi MatsuokaTRC Massively Parallel Architecture Laboratory |
My name is Hiroshi Matsuoka and I work at the TRC Massively Parallel Architecture Laboratory. It was difficult for me to decide what to write about because this is the first time I have had such an oppoitunity. Since I am a hardware researcher and not a professional writer, I have decided to be straightforward and begin with a description of my job.
At the Massively Parallel Architecture Laboratory, we are working on a massively parallel general- purpose computer that is using a new theoretical approach. This new computer will become the backbone technology for next-generation data processing. The greatest feature of this computer is that it can efficiently handle matters that conventional computers cannot effectively process. Let me briefly explain the characteristics of the computer from a technological perspective, using numelical computations for examples.
The calculation of the product of matrices can be called as the key technique of numerical calculation. The calculation is extensively used in many applications, including numerical simulation for aircraft design. In the calculation, each element of the resultant matrix can be obtained as the sum of products of elements in corresponding rows and columns of the two original matrices. The calculation of the product of matrices has a noteiworthy characteristic:it is so parallel that efficient calculation can be conducted virtually on any parallel computer of any structure because a calculation of matrix elements can be separately carried out.
In addition, the sum of the products of the resultant matrix's elements can be calculated in a sequential manner:thus the timing when the data for the calculation of j-th element will be needed can be predicted at the time of the calculation of i-th element. Since the timing for when data will be required is known beforehand, a speed-up of calculation can be achieved by obtaining that data in advance.
I'm afraid I'm getting too big for my own boots because I'm not an expeit in numerical calculation, but it can be said that what matters is the calculation speed-up can be achieved without the kind of complex programming that has been picked as parallel problems (or problems that can be processed in a parallel way).
Of course, the efficient calculation of the product of matrices is not the only qualification for massively parallel computers. The more complex the structure of data, the more difficult the prediction of timing when the data will be needed; hence, higher calculation speed will become more difficult to achieve. Our massively parallel computer is designed to avoid these problems by using original technology to reduce access time for obtaining the data necessary for calculation.
As a researcher, I am dedicated to the development of new hardware. I hope that our new computer will pave the way to wider applications and research subjects that have been considered out of the scope of parallel processing.