WIT Press


Distributed Deep Processing

Price

Free (open access)

Paper DOI

10.2495/HPC000351

Volume

23

Pages

7

Published

2000

Size

713 kb

Author(s)

R.C. Aylward

Abstract

Currently deep processing is done on high-end workstations and mainframe computers. This project explores the use of clustered digital signal processing (DSP) processors as an inexpensive alternative processing plant for deep processing tasks. In this paper an operating system is introduced and its benefits discussed. The main component of the operating system is the scheduler. The approach followed in the design is based on a tree or clusters viewed as a ring structure. This architecture reduces the problems experienced when scaling the system, Keong Loh [1]. The scheduler is distributed in the sense that any tasks arriving at one of the ports are scheduled not only by that port, but tasks are rescheduled as they migrate through the cluster until they are finally executed. Scheduling is done based on task granularity and expected time to complete as a form o

Keywords