頁籤選單縮合
題 名 | 於雲端環境下設計調適性資源分配法以改善處理績效之研究=Design Adaptive Resource Allocation Scheme to Improve the Processing Performance in the Cloud Computing |
---|---|
作 者 | 楊欣哲; 徐崇峰; | 書刊名 | 資訊管理學報 |
卷 期 | 21:3 2014.07[民103.07] |
頁 次 | 頁241-270 |
分類號 | 028 |
關鍵詞 | 雲端運算; 虛擬機器; 調適性資源分配法; 動態資源分配法; Cloud computing; VM; ARAS; VMware; DRS; |
語 文 | 中文(Chinese) |
中文摘要 | 雲端運算是將計算資源如 CPU、記憶體、磁碟和其他相關計算資源虛擬化後 的集合,其主要的服務模式含基礎設施即服務(IaaS)、平台即服務(PaaS)和軟 體即服務(SaaS)。於雲端環境中,虛擬化技術是將基礎設施封裝成虛擬機器形式 以產生量化的計算資源,然後依據實際環境工作排程資源需求的不同加以分配。 因此,虛擬化計算資源能提供合適的動態資源分配與管理是目前有待解決的重要 議題之一。雲端運算運行於VMware 架構下使用動態資源分配法(DRS) 建立根 資源池,以設定計算資源至每一虛擬機器上。當建立根資源池時所設置的計算資 源在系統處理時大於我們的預期值而無法適當利用虛擬機器資源,所以本文提出 可動態地調整虛擬機器之計算資源的機制,稱之為調適性資源分配法(ARAS)。 當雲端環境之虛擬機器高負載情況時,ARAS 機制以不中斷的服務方式即時動 態地增加每一虛擬機器之CPU 和記憶體單元以達到較有效率地調整其計算資源。 於雲端環境的批次處理之排程中使用Hyper pi 執行模擬實驗並提出四種關鍵績效 指標進行ARAS 和VMware DRS 之績效比較與分析,實驗結果證明ARAS 相對於 VMware DRS 能有效地減少約5%~26% 處理時間、提升約7.2%~10% CPU 使用率 和使用量及提升記憶體讀/寫、快取能力2%~20%。除使用Hyper pi 外,我們亦 使用另一套雲端績效監控軟體New Relic 整合至雲端應用系統中,來模擬並比較分 析ARAS 與DRS 的績效表現差異。執行New Relic 模擬實驗結果,ARAS 之CPU 整體的處理績效與記憶體使用率表現皆比DRS 更好外,我們也分析Throughput 的 績效指標,ARAS 仍比DRS 表現較好,提升了大約14.7%的效率。總之,ARAS 提供較適當的分配排程給虛擬機器以及有效地從根資源池調整其計算資源,整體 實驗結果證明ARAS 比VMware DRS 有較高的計算資源使用率與較佳的處理時間 並提升雲端應用系統的處理績效。 |
英文摘要 | Purpose: Cloud computing is a collection of virtualized computing resources such as CPU, memory, and other related resources, there is a need to support resources load balancing management. This paper aims to propose the ARAS (Adaptive Resource Allocation Scheme) to improve the VMWare DRS. Also, the ARAS can on-line monitor, tune, and adjust computing resources for enhancing the processing performance of cloud applications under cloud computing environment. Design/Methodology/Approach: When cloud computing is running on the VMware, it utilizes DRS to create a pool for setting computing resources to each VM that then activate these VMs while creating a resource pool reserved for which child resource may be larger than their needs. Hence, this paper proposes ARAS scheme to dynamically monitor and adjust computing resources under heavy loads on the cloud. Findings: We perform simulations using Hyper pi to examine that the task scheduling in cloud environments is to be compared between ARAS and DRS via four KPIs. Hence, simulation results via Hyper pi show that the ARAS can reduce processing time about 5%~26%, enhance CPU utilization about 7.2%~10%, and increase memory Read/Write and caching ability about 2%~20%. Except using Hyper pi tool, we utilize a cloud performance simulation tool: New Relic and integrate it into cloud applications to simulate and compare the performance between ARAS and DRS. The simulation resultsvia New Relic indicate that the ARAS can obtain better CPU, memory and caching utilizations, a higher throughput about 14.7% than DRS. Research limitations/implications: Our simulation environment is only performed on the two physical machines: Server1 and Server2. Also, further research is recommended to evaluate more than two physical machines and also support live migration in virtualized cloud computing. Practical/implications: The virtualization can encapsulate the infrastructure into VMs and create quantitative resources under cloud computing. Also, according to different resource needs of the actual environment, the ARAS method is more suitable for resource management to solve the loading of virtualized computing resources. Originality/Value: The proposed ARAS is a more effective and efficient resource-management method than the VMWare DRS; it can on-line monitor and tune the computing resources on the VM. Also, the ARAS scheme can obtain better processing time and higher computing-resource utilization than DRS to enhance the processing performance of cloud applications. |
本系統中英文摘要資訊取自各篇刊載內容。