Energy-Conscious Memory Allocation and Deallocation for Pointer-Intensive Applications

Abstract

A multi-bank memory architecture is composed of multiple memory banks, each of which can be energy-managed independently. In this paper, we present a set of strategies for reducing energy consumption in a multi-bank memory architecture using energy-conscious dynamic memory allocation/deallocation. Applications that make dynamic memory allocations are used very frequently in mobile computing/networking area. Our strategies focus on such applications and try to cluster dynamically created data with temporal affinity in the physical address space such that the data occupy a small number of memory banks. The remaining banks can be shut off, saving energy. All of our strategies have been implemented and tested using an in-house energy simulator and an application suite that consists of nine pointer-intensive real-life applications. Our results show that all the strategies considered in this paper save energy (e.g., our user-initiated strategy saves 49% leakage energy on the average). The results also indicate that the best savings are obtained when energy-aware memory allocation/deallocation is combined with automatic data migration.