Energy-Aware Dynamic Resource Scheduling for Distributed Multimedia Computing on Wirelessly Networked Handheld Devices

摘要

We present distributed multimedia computing on wirelessly networked handheld devices (WNHHD), where multimedia data are massively computed when communicated between the server and the handheld device (HHD). We present that, to minimize battery utilization on the HHD, a tradeoff often desires to be made between the energy utilized by computation versus that by data communication. We extend an energy cost model and present a module partition and allocation scheme to distribute the data-processing liability between the server and the HHD. Further minimization of battery utilization on the HHD is achieved tlirough minimization of computation on HMD device. Intensive experiments/simulations are performed on multiple chain-like structured programs. Our approach shows significant saving in energy over the conventional single server mode of processing data locally on the HHD. The stream of these low end-systems HHDs and high-speed networks augmented to a new class of distributed multimedia applications. These applications are naturally dynamic and highly demanding on Quality of Service (QoS). The QoS guarantee of multimedia applications is a core problem in resource management of dynamic distributed multimedia systems. An effective way to achieve resource management is to link up these end-users machines with multiple potential heterogeneous idle or under-loaded servers and allowing the multiple servers with idle retrieval bandwidth to help out end-users machines that are low power or temporarily overloaded (bottleneck). The multiple servers approach provides scalability and availability compared to single server. In this paper we design and implement new server-client architecture for the multimedia applications along-with a new algorithm. Rigorous simulation for the algorithm has been performed by varying both the size of requests (sessions) and the number of potential servers. The results show that our approach provides 1) battery management through resource scheduling 2) scalability and availability in system. The main objectives are to operate these end-systems HHDs on these chains of modules that eventually lead to the least execution time and power utilization. The results show that average battery utilization of HHDs in case of 1, 2, 3, and 4 servers are approximately 54.80%, 43.30%, 38.25%, and 32.50% respectively. The results also support that our system is scalable and fault tolerant.