Object cache : a dynamically configurable and energy efficient cache architecture for embedded systems

Abstract

Object-oriented programming languages provide a rich set of features that provide significant software engineering benefits. The increased productivity provided by these features comes at a justifiable cost of complexity in the runtime environment. This complexity leads to reduced performance and increased energy consumption of the platform as well as the programs. To alleviate the problem of increased energy consumption in embedded system architectures that typically support runtime environments for object oriented programs, this thesis proposes an energy-efficient cache architecture that can have a significant impact on the overall system energy consumption. The proposed Object-cache architecture consists of a data cache (reduced in size) and an additional small cache structure that caches only the objects called the Objectcache. A high degree of temporal locality among a large number of short lived objects ensures good performance of such an architecture. At the same time, the reduced active cache size, at times accompanied by minor improvements in performance, leads to a significant improvement in the energy behavior of programs. Using applications from SPECjvm98 benchmark and a cycle accurate simulation, the Object-cache architecture is shown to reduce average data cache energy consumption by up to 35.95% and overall system energy consumption by up to 12.4%. Statically configuring the configuration of the Object-cache is also explored. Object cache architecture is intended to be used in all systems that support runtime systems for object oriented programs. Potentially, a variety of object oriented programs would be required to run on the same architecture. Each such program would have different behavioral characteristics. Such varied behavioral characteristics of the programs would mean that each program would have different optimal Object cache configuration. As the object cache architecture is used for an increasingly wide variety of applications, a “one-size-fits-air design philosophy will be inadequate. Thus we propose an adaptive object cache architecture and a reconfiguration algorithm for dynamically reconfiguring the size and associativities of the Object-cache such that it obtains the best possible energy efficiency with the least impact on performance. Reconfiguration algorithms are proposed to be implemented on a per-phase basis. A hardware implementable energydelay- product based approximation of an ideal reconfiguration algorithm is proposed for the adaptive object cache architecture. Using applications from SPECjvm98 benchmarks and a cycle accurate simulation, the adaptive Object-cache architecture is shown to reduce average data cache energy consumption by up to 43% and also improve performance by about 1 %.