Evaluating the Self-Optimization Process of the Adaptive Memory Management Architecture Self-aware Memory

TL;DR

This paper presents a decentralized, self-optimizing memory architecture called Self-aware Memory (SaM) designed for manycore systems, demonstrating that its adaptive optimization process improves runtime efficiency despite overhead.

Contribution

It introduces the concept of ongoing decentralized self-optimization in memory management and evaluates its parameters for the first time.

Findings

Overhead of self-optimization is offset by runtime improvements.

Proper parameter tuning enhances the efficiency of the adaptive process.

Decentralized approach offers scalability and flexibility in memory management.

Abstract

With the continuously increasing integration level, manycore processor systems are likely to be the coming system structure not only in HPC but also for desktop or mobile systems. Nowadays manycore processors like Tilera TILE, KALRAY MPPA or Intel SCC combine a rising number of cores in a tiled architecture and are mainly designed for high performance applications with focus on direct inter-core communication. The current architectures have limitations by central or sparse components like memory controllers, memory I/O or inflexible memory management. In the future highly dynamic workloads with multiple concurrently running applications, changing I/O characteristics and a not predictable memory usage have to be utilized on these manycore systems. Consequently the memory management has to become more flexible and distributed in nature and adaptive mechanisms and system structures are needed. With Self-aware Memory (SaM), a decentralized, scalable and autonomous self-optimizing memory architecture is developed. This adaptive memory management can achieve higher flexibility and an easy usage of memory. In this paper the concept of an ongoing decentralized self-optimization is introduced and the evaluation of its various parameters is presented. The results show that the overhead of the decentralized optimization process is amortized by the optimized runtime using the appropriate parameter settings.