异构计算资源调度器的设计与实现毕业论文

 2021-11-23 09:11

论文总字数:16501字

摘 要

现如今,随着计算机技术和其他各领域业务的飞速发展,众多产业也正式迈入了信息时代。在这样一个背景下,各行各业内部和彼此之间的信息交流都随着业务的拓展和科技的进步都水涨船高,呈现出一个爆发增长的态势,数据量和计算量与日俱增。在这个信息爆炸的时代,传统的以CPU为核心的计算系统已然不能满足这样突飞猛进计算需求,而新型的计算系统诸如量子计算、神经网络计算等却还存在诸多难点需要解决,无法投入实际使用。目前,异构计算系统得益于其丰富多样的计算资源和对这些资源的高效利用,在高性能计算领域备受青睐。对于异构计算系统来说,多类别的计算资源是其根本,但对这些资源的合理高效地利用也是不可或缺。更高效率的任务调度是对异构计算系统的计算能力进一步拔高,对于信息时代的计算需求有着重要的意义。

Kubenetes作为一个开源的容器集群管理系统,深受业界认可。通过应用容器化,它可以实现多种多样的系统功能,其中就包括任务调度的功能。Kubernetes最核心的功能就是任务调度,而其实现这一功能的组件就是资源调度器——Scheduler。Scheduler很灵活,是以插件的形式接入集群中,这意味着我们可以去进行定制化和二次开发,即编写一个自定义的Kubernetes调度器,用以接入集群中,按照自己设定的方式进行任务调度。

首先对异构计算的概念、原理和分类进行了简要的阐述。然后介绍了Kubernetes的一些概念,对Kubernetes的系统架构进行了分析。研究Kubernetes核心组件Scheduler的运行方式和调度流程,以及如何编写自定义调度器。

关键词:异构计算;Kubernetes;任务调度;

Abstract

Nowadays, with the rapid development of computer technology and other fields of business, many industries have officially entered the information age. Under such a background, the information exchange within and between all walks of life is rising with the development of business and the progress of science and technology, showing a trend of explosive growth, and the amount of data and calculation is increasing day by day. In this era of information explosion, the traditional CPU based computing system has been unable to meet such rapid computing needs, while new computing systems such as quantum computing and neural network computing still have many difficulties to solve and can not be put into practical use. At present, heterogeneous computing systems are favored in the field of high-performance computing due to their rich and diverse computing resources and efficient utilization of these resources. For heterogeneous computing systems, multi class computing resources are fundamental, but the rational and efficient use of these resources is also indispensable. More efficient task scheduling is to further enhance the computing power of heterogeneous computing systems, which is of great significance to the computing needs of the information age.

As an open source container cluster management system, kubenetes is recognized by the industry. Through the application of containerization, it can achieve a variety of system functions, including task scheduling. The core function of kubernetes is task scheduling, and its component to realize this function is resource scheduler scheduler. The scheduler is very flexible. It is connected to the cluster in the form of plug-ins, which means that we can carry out customization and secondary development, that is to write a custom kubernetes scheduler to access the cluster and schedule tasks in the way set by ourselves.

Firstly, the concept, principle and classification of heterogeneous computing are briefly described. Then it introduces some concepts of kubernetes and analyzes the system architecture of kubernetes. This paper studies the operation mode and scheduling process of scheduler, which is the core component of kubernetes, and how to write a custom scheduler.

Key Words:heterogeneous computing;Kubernetes;task scheduling

目 录

摘要……………………………………………………………………………………I

Abract…………………………………………………………………………………I

第一章 绪论…………………………………………………………………………1

1.1 背景和意义…………………………………………………………………1

1.2 国内外研究现状……………………………………………………………2

1.3 主要工作和结构安排………………………………………………………2

1.3.1 主要工作……………………………………………………………2

1.3.2 结构安排……………………………………………………………3

第二章 异构计算系统和任务调度…………………………………………………3

2.1 异构计算的概念和原理……………………………………………………3

2.2 异构计算的分类……………………………………………………………4

2.3 异构计算的流程……………………………………………………………5

2.4 本章小结……………………………………………………………………5

第三章 集群环境下异构计算资源的调度…………………………………………5

3.1 Docker概述……………………………………………………………5

3.2 Kubernetes概述……………………………………………………………6

3.3 Kubernetes 系统架构………………………………………………………7

3.4 调度流程……………………………………………………………………8

3.5 自定义调度器……………………………………………………………10

3.6 本章小结…………………………………………………………………10

第四章 总结………………………………………………………………………11

致谢…………………………………………………………………………………12

参考文献……………………………………………………………………………13

绪论

背景和意义

近些年,中国以让世界震惊的速度在发展,经济和科技的进步肉眼可见,人们的生活也产生了翻天覆地的变化。日新月异的科学技术和迅猛发展的经济让各行各业都拓宽了业务,也产生了许许多多的新行业。这些领域或多或少都与互联网技术接轨,利用互联网的优势,对传统行业进行改革升级,形成“互联网 传统行业”的新型经济形态。在“互联网 ”这一理念的推动下,社会自然而然地进入了一个与信息和互联网息息相关的信息时代。

随着信息时代的到来,各个领域的技术和业务得到了突飞猛进的发展,也正是如此,信息量和计算需求也随之水涨船高。而CPU作为传统的处理器,计算能力和处理能力又不足以应对这种爆发式增长的信息量和计算需求。作为一款集成了指令运算和数值运算的通用处理器,CPU以指令计算为主,数值计算为辅,注重指令执行的并行性,对于处理复杂的指令操作很擅长,适用于逻辑密集的控制型任务,但在计算密集型任务的处理上并没有那么得心应手。然而由于集成电路制造工艺遭遇瓶颈,摩尔定律已经失效,再加上之前提到的信息时代对庞大信息量和计算量的处理能力的要求,以CPU为核心的传统计算系统已经远远不能满足各领域的计算需求,因此,在这个“后摩尔时代”,需要一种新型的计算系统来承担这项任务。

请支付后下载全文,论文总字数:16501字

您需要先支付 20元 才能查看全部内容!立即支付

该课题毕业论文、开题报告、外文翻译、程序设计、图纸设计等资料可联系客服协助查找,微信号:bysjorg 、QQ号:3236353895;