嵌入式多媒体流计算的自适应机制与跨层优化

项目名称： 嵌入式多媒体流计算的自适应机制与跨层优化

项目编号： No.60873029

项目类型： 面上项目

立项/批准年度： 2009

项目学科： 金属学与金属工艺

项目作者： 郭红星

作者单位： 华中科技大学

项目金额： 30万元

中文摘要： 由于目前常用的嵌入式多媒体流计算系统缺乏动态优化机制，系统不能根据需求变化定制所需计算资源，导致资源利用效率不高。本项目利用新型流处理器增加的硬件可伸缩功能，研究嵌入式多媒体流计算的自适应优化问题，内容包括用于联合优化的系统分析模型，支持双向信息传递的跨层结构和资源受限条件下的系统优化方法。创新性地结合转移函数法和状态变量法建立系统分析模型，提出了多媒体流计算复杂度的建模与动态预测方法；并根据多媒体流计算的软实时性，设计双向跨层信息传递结构以替代传统的单向信息传递结构，实现计算资源和服务质量的联合优化。通过预测和调度主动利用流计算任务的统计特性，均衡计算负载；通过调节应用层服务质量，以应对硬件可配置粒度限制，提高系统资源利用效率。通过数据布局、线程级并行和多任务调度优化，实现对共享资源的合理分配与高效利用。以跨层适应方式，优先传输和处理重要程度更高的数据包，从而获得更优的服务质量。这种计算模式具有适应外在需求变化动态优化自身组织结构的能力，可节省系统能耗。研究结果为新一代节能型嵌入式多媒体系统的优化设计提供了新思路，对丰富和发展嵌入式流计算的学科理论有重要意义，具有广泛的应用前景。

中文关键词： 嵌入式系统；流计算；多媒体系统；自适应；跨层优化

英文摘要： Current embedded systems for multimedia stream computing are usually lack of dynamic optimization mechanisms, although more and more scalable hardware functions are added to the new type of stream processors. So, the embedded systems are not capable of tailoring appropriate resources to actual requirements, which leads to a relative low resource utilization efficiency. This project investigates adaptive optimization schemes for the embedded multimedia stream computing, which includes system models for joint optimizing, cross-layer structures for bidirectional passing information and optimization methods under resource constraint. An innovative system model, which combines the transfer function method and the state variable method, will be set up. On the basis of this model, a linear model-based method is proposed to predict decoding computational complexity adaptively online. Aim at soft real-time characteristics of multimedia stream computing, a cross-layer structures for bidirectional passing information, which substitutes the conventional unidirectional structures for information passing, is used to implement joint optimization of computing resources and quality of services. Based on predicting and scheduling, loads are balanced by forwardly making use of the statistical features of stream computing tasks. By scaling of Quality-of-Service(QoS) of applications to deal with the configurable granularity constraint of system resources, the utilization efficiency of the resources can be further improved. Through optimization of data layout, thread level parallelism and task scheduling, the share resources are allocated appropriately and used efficiently. Better QoS can be achieved by transmitting and processing more important data packets first in a cross layer adaption framework. This kind of computation mode has the capability of dynamical self-tuning to fit extrinsic requirements, which is a key for saving energy consumption of embedded systems. Results of the project will provide new approaches for the optimization design of next generation energy-efficient embedded multimedia systems. It is also very important for enrichment and development of subject theories of the embedded stream computing and shows extensive and promising applications in embedded systems.

英文关键词： Embedded systems; stream computing; multimedia systems; adaptation; cross-layer optimization