随着行业应用的飞速扩张,数据中心以及云等日益成为主流服务平台.高性能的片上多核系统也随之成为重要的可分配资源之一.然而,在对多用户提供服务(并发执行、并置执行)时,其所固有的共享资源会引发严重的并发性能下降.在此背景下,多核系统的性能以及资源利用率问题成为研究热点.软件调度策略作为一种性价比较高的调节手段对于缓解资源冲突一直行之有效,然而,硬件技术的变迁对其调节的效力将产生一定影响.首先从片上多核结构关键技术入手,对共享资源的变化进行了详细阐述,在此基础上,对当前主流应用领域中两种不同类型的软件调度策略进行介绍和分析.在总结现有策略的局限性以及面临的新挑战的同时,对未来的研究趋势作了展望.
With the fast swelling of industry applications, new platforms, such as clusters, datacenter and cloud computing, are becoming mainstream service environments. Meanwhile, high performance chip multiprocessor (CMP) systems begin to play a role on these commercial service platforms as major allocable resources. However, shared resources of CMP architectures would lead to severe co run performance degradation on concurrent user applications. This problem, if left untouched, would lower the resource utility even users have less demands on the quality of services (QoS). Shared resources are attracting more and more concerns. Proper software scheduling policies keep being effective to the improvement of co-run performance. The paper summarizes the development of key techniques on CMP, variations of shared resources, which are followed with a detailed presentation and analysis for the representative schedulers. Limitations of the state-of-the-art scheduling policies, new challenges are then discussed and prospects for the research trends in the future are presented at~ the end of the paper.