An Effective Early Multi-core System Shared Cache Design Method Based on Reuse-distance Analysis

TL;DR

This paper introduces a novel reuse-distance analysis method for optimizing multi-core shared-cache design early in the development process, achieving high accuracy with minimal error in cache miss estimation.

Contribution

It presents an efficient approach to derive aggregated reuse-distance histograms for concurrent applications, enabling accurate cache design optimization without hardware dependency.

Findings

Average error rate of cache miss estimation is less than 2.4%.

The method accurately predicts shared-cache contention.

Optimal cache configurations can be identified early in design.

Abstract

In this paper, we proposed an effective and efficient multi-core shared-cache design optimization approach based on reuse-distance analysis of the data traces of target applications. Since data traces are independent of system hardware architectures, a designer can easily compute the best cache design at the early system design phase using our approach. We devise a very efficient and yet accurate method to derive the aggregated reuse-distance histograms of concurrent applications for accurate cache performance analysis and optimization. Essentially, the actual shared-cache contention results of concurrent applications are embedded in the aggregated reuse-distance histograms and therefore the approach is very effective. The experimental results show that the average error rate of shared-cache miss-count estimations of our approach is less than 2.4%. Using a simple scanning search method, one can easily determine the true optimal cache configurations at the early system design phase.