Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/98250
標題: 實現多用戶效能公平之記憶體快取動態調整方法
A Page Cache Management Scheme for Performance Fairness
作者: 廖健能
Chien-Neng Liao
關鍵字: 雲端服務
機械式硬碟
固態硬碟
記憶體快取
公平的效能
cloud computing
SSD
mechanical hard disk
memory caching space
performance fairness
引用: [1] S. Mittal, 'A survey of techniques for cache partitioning in multicore processors,' ACM Comput. Surveys, vol. 50, no. 2, pp. 1–27, May 2017. [2] Q. Pu, H. Li, M. Zaharia, A. Ghodsi, and I. Stoica. Fairride: Near-optimal, fair cache sharing. In NSDI, 2016. [3] C. M. Patrick, R. Garg, S. W. Son, and M. Kandemir. Improving i/o performance using soft-qos-based dynamic storage cache partitioning. In Cluster Computing and Workshops, 2009. CLUSTER'09. IEEE International Conference on, pages 1–10. IEEE, 2009. [4] D.-W. Chang, H.-H. Chen and W.-J. Su, 'VSSD: Performance Isolation in a Solid-State Drive,' in the 12th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid), 2012. [5] J. S. Bucy, J. Schindler, S. W. Schlosser, and G. R. Ganger. The disksim simulation environment version 4.0 reference manual (cmu-pdl-08-101). Parallel Data Laboratory, page 26, 2008. [6] C. Gniady, A. R. Butt, and Y. C. Hu, 'Program-counter-based pattern classification in buffer caching,' in OSDI, 2004, pp. 27–27. [7] AccuSim, http://cobweb.ecn.purdue.edu/~ychu/accusim/. [8] Micron Technology Inc., https://www.micron.com/products/dram/ddr3-sdram.
摘要: 現今網路已發展出各式不同的雲端服務及應用,對於雲端服務供應商而言,必須確保使用者能享有穩定且公平的服務品質。另外,資料系統儲存裝置被視為影響整體服務時間的關鍵,也被視為系統效能的最大瓶頸。過去由於固態硬碟的容量較小、價格較高且有壽命限制,以致於無法完全取代傳統機械式硬碟,但近年來固態硬碟技術發展成熟,固態硬碟已逐漸替代傳統硬碟為主要儲存裝置,將大大改善系統整體效能。 在資源共享的雲端環境中,多位使用者競爭共有的記憶體快取空間,由於使用者彼此存取特性的不同,造成了使用者記憶體快取空間分配不均,進而產生使用者效能不公平的情況。因此,本篇論文基於固態硬碟儲存裝置,探討如何公平的分配記憶體快取空間,使得多位使用者能夠獲得公平的效能(Performance fairness)。在此系統中,將會給予每位使用者一個權重值,並根據每位使用者所獲的權重值作為系統資源(記憶體快取空間、使用儲存裝置的時間)分配的標準,並且使用Credit-Based I/O Scheduler來對使用者的 I/O請求收取適當的服務成本,再藉由記錄使用者每週期Read/Write Request的數量和在記憶體快取的Hit ratio,做為下一週期動態調整使用者快取空間的依據,使每位使用者獲得公平的效能,以達成Performance Fairness的目標。
Nowadays, as cloud computing has grown in popularity, service providers have developed a variety of different cloud services and applications. For service providers, it is necessary to ensure that users have stable and fair service quality. In addition, storage devices have been recognized as the key of affecting the overall service time and thus considered as the biggest bottleneck of system performance. In the past, due to the small capacity, high price and wear endurance of SSDs, it was impossible to completely replace traditional mechanical hard disks. However, in recent years, solid-state hard disk technology has significant improved. Thus, solid-state hard disks have gradually replaced traditional mechanical hard disks, which greatly improves the overall performance of the computer system. In the cloud environment, resources are shared among users, multiple users would compete for the shared page cache space. Due to the difference user access characteristics, the memory caching space is allocated unevenly, which in turn causes users obtaining unfair performance. Therefore, on the basis on solid-state drives as the storage devices, this paper explores how to allocate memory cache space fairly, so that users can achieve performance fairness. In this system, each user will be given a weight value. The weight value of each user will be used as the standard for system resource allocation. We also use the credit-based I/O Scheduler to charge users for the appropriate service cost of I/O requests. Besides, we record the number of read/write requests for each in each period and calculates the hit ratio of different cache sizes for dynamically adjusting the user cache space in the next period, and finally, the goal of performance fairness can be achieved.
URI: http://hdl.handle.net/11455/98250
文章公開時間: 2019-01-30
Appears in Collections:資訊科學與工程學系所

文件中的檔案:

取得全文請前往華藝線上圖書館

Show full item record
 
TAIR Related Article
 
Citations:


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.