基于温度感知的多核系统任务调度策略研究

发布时间：2018-05-13 02:03

本文选题：多核 + 温度感知　；参考：《武汉理工大学》2013年硕士论文

【摘要】：在多核(Multi-core)处理器蓬勃发展的今天,以双核,四核处理器为代表的多核产品越来越深入到人们的生活中。人们在享受多核产品带来的便利高效时,也从未停止过提高其性能的步伐。然而,随着处理器内部晶体管数目的增加和集成度的提高,带来了诸如功耗增加,发热增大等新的问题。因此,近年来,如何降低多核系统中处理器的温度,以提高处理器性能和资源利用率成为新的研究热点。目前的研究方向主要分为硬件层面和软件层面。采用硬件技术能在一定程度上降低处理器温度,但牺牲了处理器部分性能。因而,近年来的研究重点主要转向软件方面,期望通过合理的启发式思想和有效的任务调度策略来达到降低处理器温度的目的。基于目前通用的温度模型和功耗模型,本文首先从温度感知的角度出发,通过分析和比较了现有的几种启发思想之后,基于最冷核的策略,并结合热任务优先的调度思想,本文提出了MCP_HOT算法。该算法能够很好的感知处理器温度,通过不断的动态调整,不仅能保证任务有效的满足截止期的需求,同时还能有效缓解单个处理器的温度压力,减少热点产生。接着,本文从提高处理器的任务利用率和负载均衡的角度出发,在对现有的三种经典的分配算法进行了深入分析和比较后,提出根据任务集属于重载还是轻载任务确定不同的分配因子,有助于提高处理器的利用率。同时,本文结合DVFS技术中的空闲时间回收策略,提出了LB_DVFS算法。该算法的设计目标是提高处理器的任务利用率和负载均衡,通过DVFS技术有效的降低处理器功耗,从而降低处理器温度。最后,基于上述两种算法,在仿真实验中,利用HotSpot温度-功率模型,对本文提出的两种算法进行了论证。实验结果证明,第一种算法能够有效地降低处理器峰值温度,而第二种算法一方面能有效提高处理器上的任务利用率,另一个方面也能有效降低处理器功耗和温度。
[Abstract]:With the rapid development of multi-core processors, the multi-core products, represented by dual-core and four-core processors, are becoming more and more popular in people's lives. People have never stopped improving their performance while enjoying the convenience and efficiency of multi-nuclear products. However, with the increase in the number of transistors and the increase in integration, new problems such as increased power consumption and increased heat are brought about. Therefore, in recent years, how to reduce the processor temperature in multi-core systems to improve processor performance and resource utilization has become a new research hotspot. The current research direction is mainly divided into hardware level and software level. Using hardware technology can reduce processor temperature to a certain extent, but at the expense of part of processor performance. Therefore, in recent years, the research focus is mainly on software, and it is expected to reduce the processor temperature through reasonable heuristic and effective task scheduling strategy. Based on the current general temperature model and power consumption model, this paper first analyzes and compares several existing heuristic ideas from the point of view of temperature perception, then combines the strategy based on the coldest kernel and the scheduling idea of thermal task first. In this paper, MCP_HOT algorithm is proposed. The algorithm can sense the processor temperature well, through continuous dynamic adjustment, can not only ensure that the task can effectively meet the deadline requirements, but also can effectively alleviate the temperature pressure of a single processor, reduce the hot spot generation. Then, in order to improve the task utilization and load balance of the processor, this paper analyzes and compares the existing three classical allocation algorithms. It is suggested that different allocation factors can be determined according to whether the task set is heavy or light, which is helpful to improve the utilization of processors. At the same time, combined with the idle time recovery strategy in DVFS technology, the LB_DVFS algorithm is proposed. The design goal of the algorithm is to improve the task utilization and load balance of the processor, and to reduce the processor power consumption and processor temperature effectively through DVFS technology. Finally, based on the above two algorithms, the HotSpot temperature-power model is used to demonstrate the two algorithms proposed in this paper. Experimental results show that the first algorithm can effectively reduce the processor peak temperature, while the second algorithm can effectively improve the task utilization on the processor, the other side can effectively reduce the processor power consumption and temperature.
【学位授予单位】：武汉理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP332;TP301.6

【参考文献】