Energy-aware resource management for heterogeneous systems

Energy-aware resource management for heterogeneous systems

Abstract

Nowadays computers, be they personal or a node contained in a multi machine environment, can contain different kinds of processing units. A common example is the personal computer that nowadays always includes a CPU and a GPU, both capable of executing code, sometimes even in the same integrated circuit package. These are the so called heterogeneous systems. It’s important to be aware that the various processing units aren’t equal, for instance CPUs are very different from GPUs. This raises a problem, since not every task can be executed in all processing units. To solve this problem a new task scheduling algorithm was developed with the aid of SimDag from the SimGrid toolkit. This algorithm uses a DAG (directed acyclic graph) to aid the scheduling of different tasks, be they from a single application or from various different applications. The algorithm is based on the HEFT scheduling algorithm, a greedy algorithm with a short execution time, developed by Topcuoglu et al. This new algorithm is aware of the different processing units and of the different performance/power levels. This solves the problem of not all tasks being able to be executed in all processing units. Since previous studies show that reducing the CPU clock speed on DVFS (dynamic voltage frequency scaling) CPUs can reduce the energy spent by the CPU while executing various tasks with little increase in runtime. Various tests were made to obtain the power rating of a test CPU while operating on different performance levels. With this it was possible to obtain performance and power information on the power states, this information is then later used by the algorithm in order to find the optimal performance/power ratio. The algorithm main objective is to spend the least amount of energy possible, in contrast to the HEFT goal that is to execute tasks as fast as possible. The algorithm behavior can be modified by changing the minimum power state that the processing units should run or by changing the goal. Two goals are provided, the EFT (earliest finish time) from the original HEFT algorithm and the LEC (least energy cost). Both goals are affected by the defined minimum power state. Using this new algorithm it was possible to reduce total energy spent some times at the cost of increased runtime.

BUY THIS TOPIC ON WHATSAPP

Related Project Materials

ABSTRACT 

The purpose of this project is to examine the level of acceptability between the self-assessment scheme a...

ABSTRACT

This work titled, “Implication of Local Government Autonomy on Rural Development in Ebonyi State” A stu...