Font Size: a A A

Disk IO QoS Control On Virtualized Cluster

Posted on:2014-01-23Degree:MasterType:Thesis
Country:ChinaCandidate:Y B WuFull Text:PDF
GTID:2248330392961083Subject:Software engineering
Abstract/Summary:PDF Full Text Request
Cloud computing is now regarded as another great technical revolution after pre-vious Client/Server architecture revolution in computer science area. In order to meetdiferent demands of cloud users, a vast resource cluster must be hosted by cloud com-puting system. Therefore, cloud resource management leads a big challenge in thisarea. Tasks of cloud computing resource management are mainly composed of data s-torageandresourcemanagementthatarebothrelatedtothediskI/Oresource. DiskI/Oresource is the important one of all resources under cloud computing environment. Anefective disk I/O QoS (Quality of Service) system will gain big proft for cloud serviceproviders, cloud users and end users. There are many factors to consider while design-ing and implementing a disk I/O QoS scheduling system, such as high performance,fexible administration, compatibility, etc. The implementation of disk I/O resourcescheduling policy for virtualized storage and the QoS guarantee of disk I/O becomeimportant and practical research issues.The research subject comes from the National863plan program. The programhopes to implement an efective disk fow control design with virtualization and traf-fc control techniques. This design will implement the trafc control of specifc unitand guarantee the high availability of system resources. It also makes sure that cloudservice providers provide users with high quality and stable data fow programs. Toachieve this goal, this paper proposes a disk I/O resource management plan based onvirtualization and QoS control techniques, which can efectively manage and controldisk I/O trafc. This proposal includes high performance by providing several kinds ofservices and high availability of administration. The evaluations of QoS module in thesystem testing environment will be considered as experimental validation of providinga highly reliable guarantee for system online running.The paper’s main achievements consist of three parts: 1. Design and implement disk I/O QoS control module. With the combined mod-el of weighted fair queue algorithm and leaky bucket algorithm, two disk schedulingpolicies are implemented. One is throttle policy used as the limitation of system diskresources and another is weight policy used to confgure the priority of diferent re-quests. Device–mapper technique is used to implement the module as an independentvirtual block layer so as to reduce the impact of QoS module introduction on systemperformance.2. Implement QoS Administration interfaces. Relevant scripts of APIs need re-vising to implement QoS administration in order to improve the availability of systemintegration and parameter administration. The performance test gives an evaluation ofimpact on system integration of QoS module.3. Analyze optimization factors of disk I/O. Test results of system disk I/O il-lustrate that factors, afecting the system disk I/O, include the size of IO requests andnumber of concurrent IO requests. The control of these factors can improve the ef-ciency of disk I/O and optimize system performance.This paper proposes a disk I/O QoS control module based on virtualization tech-nology, after analyzing and researching on the current situation of disk resource man-agement techniques in a virtualized environment. The paper selects a proper designplan and implements it with device–mapper technique. In order to make it easy to ad-min the module, system integration design and implementation details of disk QoS aredescribed in the paper. The evaluations show that the module with whole functionsof disk I/O QoS can provide diferent kinds of I/O scheduling policies and quality ofservices for users to guarantee system efciency. The strategies can limit disk I/O re-sources around or below certain values with fuctuations not exceeding3%of overall,which avoids the bursting fow. The results imply that the loading of our QoS moduleonly leads to performance efects within2%on our testbed.
Keywords/Search Tags:Cloud computing, Resourcemanagement, Diskschedul-ing, Quality of service
PDF Full Text Request
Related items