- 學位論文
改善並行虛擬機器排程演算法之研究
Improving Scheduling for Concurrent Virtual Machines
摘要
在虛擬化平台Xen Hypervisor上，有許多組態設定，像是Virtual CPU(VCPU)、Virtual Memory、Scheduling等皆是用來架構虛擬機器所需要的各種組態與排程。其中Xen Hypervisor上面之CPU Scheduling的演算策略設計為我們所深入研究之項目。目的在於改善當前Xen CPU Schdeduling的排程策略Credit下，對於執行並行程式的虛擬機器(VM)，有可能造成可觀的時間浪費。 由於在並行程式中各個同階段的任務之間需要互相通信與同步，而在Credit的排程策略下，並不會根據並行程式而同時間執行，導致各任務需要花費較多的時間，在執行同步的動作。而在Tseng et al.的方法提出，針對並行程式的VCPU，附加了Turbo的優先權，在排程階段時，將同是Turbo的VCPU一同搬移到執行佇列的頂端執行。但Credit與Tseng et al.的方法皆是使用較花費時間的佇列操作方法(插入、刪除、查找)。 我們提出利用Linux kernel 2.6版本所使用的CFS(Completely Fair Scheduler)排程方法來改善Credit所產生的影響，並藉由CFS的機制來消彌並行VM與其他VM之間的差距。CFS可以更公平的分配到處理器的時間，且基於RB-Tree的概念與操作，可以讓越少被執行到的任務，有更高的優先機率被處理器執行到，而在並行VCPU的處理上，透過標記與設置等待佇列，亦達成了並行VCPU的同步問題。 在我們的實驗與比較部分，使用NAS Benchmark所提供的並行程式做實驗，在VM中執行並逐步調整VM的權重使VCPU在RB-Tree中的Virtual runtime增減，以換取PCPU挑選VCPU的機會，最後獲取執行時間來統計效能的變化。 實驗結果可以看出在Credit、Tseng et al.的方法與我們的演算方法相比較下，確實大幅減少了全域的運行時間，在並行虛擬機器下的整體效能大幅增加。
並列摘要
On the virtualization platform Xen Hypervisor, there are many configuration settings, such as Virtual CPU (VCPU), Virtual Memory, Scheduling are needed for a variety of configuration and scheduling for virtual machine architecture. The strategy design which Xen Hypervisor above the CPU Scheduling for our in-depth research project. The purpose is to improve the current under Xen CPU Schdeduling strategy Credit, for a virtual machine (VM) implementation of a concurrent application, it may result in considerable waste of CPU time. Due to need to communicate with each other with the phase synchronization between the various tasks in a concurrent application, and in the Credit scheduling strategy, not according to the concurrent application execution while at the same time, leading to spend more execution time for task is running synchronous operation. In Tseng et al. Methods proposed for VCPU of concurrent application, attached Turbo priority when scheduling phase, along with the Turbo's VCPU move to the top of the run queue to perform. However, Credit and Tseng et al. Methods are all the more time-consuming methods of operation (insert, delete, search). We propose the use of Linux kernel 2.6 version used CFS (Completely Fair Scheduler) scheduling methods to improve the problem of Credit generated, and by mechanisms to reduce the gap between the VM and the other VM. CFS may be a more equitable distribution to the CPU time, and RB-Tree is based on the concept of operations, allowing fewer tasks to be performed, there is a higher probability of being executed by the processor priority to, and in VCPU of concurrent processing, through the mark and set the concurrent wait queue, also reached a VCPU of concurrent synchronization problems. In our experiments and comparison, using the NAS Parallel Benchmark programs offered to do the experiment, the right to perform in a VM and VM weight gradually adjusted so that Virtual runtime in RB-Tree in a decrease for the chance to pick VCPU. At last, the execution time to obtain statistical performance changes. The results can be seen in the Credit, Tseng et al. compared with our strategy indeed significantly reduce the execution time, and substantial increase in concurrent virtual machine under the overall performance.
參考文獻
延伸閱讀
- 魏永強、曾昭雄、張榮憲（2010）。改善內部預測編碼法之研究。載於遠東科技大學電機工程系（主編），智慧型系統工程應用研討會（頁436-439）。遠東科技大學電機工程系。https://doi.org/10.30176/ISC.201005.0436
- Tsai, M. H. (2013). 基於自動機理論的模型檢測演算法與工具之改善 [doctoral dissertation, National Taiwan University]. Airiti Library. https://doi.org/10.6342/NTU.2013.02165
- Huang, T. Y. (2021). 混合啟發式演算法求解非等效平行機台排程之研究 [master's thesis, National Tsing Hua University]. Airiti Library. https://www.airitilibrary.com/Article/Detail?DocID=U0016-2412202116163737
- 鄭宇軒（2021）。Improving Service Stability for Event Sourcing and CQRS Architecture〔碩士論文，國立政治大學〕。華藝線上圖書館。https://doi.org/10.6814/NCCU202101070
- Tuan, T. M. (2013). Two-tier Project and Job Scheduling for Cloud Computing Environments [master's thesis, National Chiao Tung University]. Airiti Library. https://doi.org/10.6842/NCTU.2013.00097