Dividing a Task and Executing it in Parallel on a Function-Classified Parallel Computer

OHTA MITSUHIKO, KOHNO MICHIMUNE, MAEZAWA TOSHIYUKI, ANZAI YUICHIRO

In a functionally classified parallel computer, tasks which are functionally divided are both programmed and executed in parallel in each modules. But these operations are difficult. In this paper, we propose a system that divides a user task which includes some threads into some functional modules at a run time and executes these divided tasks in parallel. In this system, a thread which frequently accesses a particular device is executed in the module which has that device. We consider that this system can reduce inter-module communications between server threads which control the device and user threads. We implemented this system into μ-PULSER, an operating system for personal robot. We evaluated some basic thread operations in this system. The result shows that the cost of creating a remote thread is 2.7 times as long as that of creating a local thread. And we also evaluated the rate of inter-module communications. The result shows that it was reduced to 50% and that the response time of a event is reduced to 8.0%.

Dividing a Task and Executing it in Parallel on a Function-Classified Parallel Computer

Bibliographic Information

Search this article

Description

Journal

References(9)*help

Details 詳細情報について

Export

Report a problem