MultiAmdahl: Optimal Resource Allocation in Heterogeneous Architectures

TL;DR

This paper extends the MultiAmdahl analytical model to optimize resource allocation for energy efficiency in heterogeneous multiprocessor architectures, including data centers, by reallocating resources based on constant system power considerations.

Contribution

It introduces an energy-aware extension of MultiAmdahl for resource allocation and provides insights on reallocating resources from general-purpose to accelerator-dominated computing.

Findings

Reducing constant system power favors reallocating resources to accelerators.

Energy optimality involves shifting resources from general-purpose units to specialized accelerators.

The approach applies to both chip architectures and data center computing.

Abstract

Future multiprocessor chips will integrate many different units, each tailored to a specific computation. When designing such a system, the chip architect must decide how to distribute limited system resources such as area, power, and energy among the computational units. We extend MultiAmdahl, an analytical optimization technique for resource allocation in heterogeneous architectures, for energy optimality under a variety of constant system power scenarios. We conclude that reduction in constant system power should be met by reallocating resources from general-purpose computing to heterogeneous accelerator-dominated computing, to keep the overall energy consumption at a minimum. We extend this conclusion to offer an intuition regarding energy-optimal resource allocation in data center computing.