請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/17159
標題: | 雲端服務營運布建及轉移排程之研究 Service Deployment and Management in Cloud Data Center |
作者: | Chin-Chia Chang 張晉嘉 |
指導教授: | 廖婉君 |
關鍵字: | 雲端資料中心網路,視訊檔案布局,虛擬主機轉移, Data center network,VM migration, |
出版年 : | 2013 |
學位: | 博士 |
摘要: | 資源池概念對於雲端資料中心而言是很重要的概念。其定義系統管理者必須根據使用者需求的改變來快速動態的調整資源量。有需求的時候才消耗資源,當需求減少時則將資源回收節省成本。由於虛擬化技術的發展,應用實體主機的方式有了很大的改變,主機群可以藉由虛擬化技術建構出一個整合的運算資源池。
我們探討在影音服務中,當需求改變時,視訊檔案的布局方法,能隨著需求變動及時增加或者回收主機,並且只需要極低的轉換成本。使系統在每個時段皆能使用最少的主機數量。 接著我們探討在雲端資料中心當中,虛擬主機轉移的排序問題。當虛擬主機離開實體主機的時候,虛擬主機的資源消耗會轉移到新的實體主機。因此,為了讓虛擬主機轉移速率上升,一個好的虛擬主機轉移排序,必須同時考量來源跟終端實體主機的剩餘資源。我們先經由數學推導,計算在單一來源跟單一終點實體主機情況下的虛擬主機排序問題。接著我們依循此排序原則,我們設計了多來源跟多終點實體主機情況下的虛擬主機移動排序問題。藉由實驗結果顯示,我們發表的SB-Mig機制可以同時擁有較短的系統轉移時間以及服務中斷時間。 Resource pooling concept is the key factor of cloud data center resource management. Virtualization technologies change the way data centers utilize their physical machine resources. Instead of using dedicated servers for each application, virtualization realize the concept of unified resource pool. For video service, cloud-based content delivery networks require on-line algorithm to adapt the video service deployment under dynamic user demand. For user demand increase, new video replicas are placed into the physical machines. While user demand decreasing will trigger the system to remove unnecessary video replicas and consolidate all video replicas to minimize the total physical machine number and reduce the operation cost. Therefore, CloudCast is proposed to solve the video replica placing problem in cloud data centers. The simulation shows that the on-line algorithm CloudCast can minimize the number of tun-time physical machines. The server number requirement is very close to lower bound under various first-fit placement policy. In the problem of service placement in data center, multiple VM migration is computed as to minimize the physical machine number for placement maintenance. In our best knowledge, it is the first paper to investigate the migration sequence of multiple VMs (video replica). For one VM migration, the VM service resource loading are shifted from the source physical machine to the destination physical machine. In source physical machine, the migration transmission data rate increase as VM leaving source. On the other hand, the migration data rate at destination physical machine is reduced. Therefore, a good migration sequence should balance the tradeoff between source and destination migration data rate as to minimize the service interruption time and system migration time. The existing method migrate all VM simultaneously, thus the migration performance is reduced. We propose SB ratio based migration sequence planning (SB-Misp) to determine the sequence of VM migration. We first derive the metric for sequence planning problem under different scenarios via mathematical induction. Based on the proposed sequence planning metric, we design algorithm to solve the sequence planning problem. The simulation shows that our proposed mechanism (SB-Misp) can reduce both service interruption time and system migration time. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/17159 |
全文授權: | 未授權 |
顯示於系所單位: | 電信工程學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-102-1.pdf 目前未授權公開取用 | 1.54 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。