Full metadata record
DC Field | Value | Language |
---|---|---|
dc.contributor.author | 賴建宇 | en_US |
dc.contributor.author | Lai, Chien-Yu | en_US |
dc.contributor.author | 黃育綸 | en_US |
dc.contributor.author | Huang, Yu-Lun | en_US |
dc.date.accessioned | 2014-12-12T02:34:34Z | - |
dc.date.available | 2014-12-12T02:34:34Z | - |
dc.date.issued | 2012 | en_US |
dc.identifier.uri | http://140.113.39.130/cdrfb3/record/nctu/#GT070060096 | en_US |
dc.identifier.uri | http://hdl.handle.net/11536/72297 | - |
dc.description.abstract | Infrastructure as a service(IaaS)是目前極為普及的一種雲端服務。現有主流的雲端服務供應商均提供多種既定的虛擬規格供使用者選擇。若使用者希望其應用程式在虛擬平台執行時,能獲得與原平台相同的效能,則這些既定的虛擬規格中,可能不會有恰好符合使用者需求的選項。所以本論文提出新的虛擬機效能評估模型(PEM),以利雲端服務供應商進行適當的虛擬資源規劃,使虛擬機的效能可以更接近使用者的各種需求。我們的效能評估模型分為三個階段:一、由雲端服務供應商與使用者個別量出雲端伺服器與使用者本機的效能參考數據;二、依使用者需求選擇適當的參數集合;三、使用多元線性迴歸函數預估最適合使用者需求的虛擬資源建議。為了滿足使用者的多項效能需求,我們的方法設計了兩種估測模式:一、最接近原平台效能的估測模式,及二、超越原平台效能的估測模式。其中,採用模式一時,所有應用程式在虛擬機上執行的整體運作效能,最接近這些應用程式在原平台上的運作效能。模式二,顧名思義,所有應用程式在虛擬機上的執行效能會優於原平台。我們最後設計了數個實驗,探討套用我們的模型後,應用程式在虛擬機與原平台上的執行效能誤差。採用模式一時,每個應用程式在虛擬機上的執行效能誤差約為8.4%到10.6%;但相較於用硬體資源比例計算,雲端服務供應商的硬體資源成本可以省下約38.5%到41.8%。採用模式二時,虛擬機上的應用程式效能皆比原平台高。至於網路的部份,應用我們的模型後,應用程式在虛擬機上的傳輸效能大多高於原平台(0.6%~14.3%)。此實驗結果證實採用我們的方法不但可以讓應用程式在虛擬機上運行時,得到使用者想要的效能表現,同時也證明我們的方法可以有效地協助雲端服務供應商有效地運用其雲端資源,並節省硬體資源與成本。 | zh_TW |
dc.description.abstract | Existing cloud providers provide multiple pre-configured VMs for selection. A subscriber can select a VM from the pre-configured options, but may not find a VM on which the application performance is similar to the original platform owned by the subscriber. In this thesis, we propose a 3-stage performance estimation model (PEM) to determine a resource plan fulfilling the performance requirements from a subscriber. Stage 1: a cloud provider and a subscriber profile their own platforms. Stage 2: the cloud provider selects proper parameters in the profiling data. Stage 3: the cloud provider applies multi-linear regression to determine a resource plan that fulfills the performance requirements specified by the subscriber. We design two modes for different performance requirements. Mode 1: the application performance of a VM is close to that in the original platform. Mode 2: the application performance of a VM exceeds that in the original platform. We conduct experiments to evaluate the accuracy of PEM. In mode 1, the performance errors of CPU applications range from 8.4% to 10.6% and the cost savings of CPU resource range from 38.5% to 41.8%. In mode 2, all performances of CPU applications exceed those in the original platform, and the performance error of network throughput ranges from 0.6% to 14.3%. From the results of our experiments, we prove that PEM can satisfy the performance requirements from subscribers and save costs of hardware resources for a cloud provider. | en_US |
dc.language.iso | en_US | en_US |
dc.subject | 虛擬機器 | zh_TW |
dc.subject | 效能評估模型 | zh_TW |
dc.subject | Performance Estimation Model | en_US |
dc.subject | VM Sizing | en_US |
dc.title | 針對虛擬機資源規劃之多階段效能評估模型 | zh_TW |
dc.title | A Multi-stage Performance Estimation Model for VM Sizing | en_US |
dc.type | Thesis | en_US |
dc.contributor.department | 電控工程研究所 | zh_TW |
Appears in Collections: | Thesis |