Scalability of Parallel Systems: Efficiency Versus Execution Time

Abstract

Ideal parallel computer systems (architecture+class-of-algorithms) are linearly scalable. They allow to solve a series of increasingly larger problems by using more processors and maintain the machine efficiency and the execution time constant. Most of the practical parallel systems are nonlinearly scalable. For these systems we must choose either a constant execution time or a constant efficiency when scaling-up the problem and the machine.