臺灣博碩士論文加值系統

由於網際網路應用的蓬勃發展，熱門網站所提供的服務也越趨多元化。這些服務通常具有即時與動態資訊等特性，因此若僅僅由一台伺服器來應付所有要求，常有可能發生過度負載的狀況。本論文將採用一種有彈性的註冊協定，使得負載平衡系統能夠因應情況，很容易的即時增加或減少提供動態網頁之伺服器。並且利用加權分配演算法分配使用者需求，讓使用者獲得公平的服務品質。我們運用多台用戶端設備在短時間內送出大量連線要求給動態網頁伺服器系統，隨機存取動態網頁。實驗結果顯示我們的系統對於提供動態網頁，確實可以根據伺服器服務能力分配負載。我們並且比較單一動態網頁伺服器與負載平衡系統之能力差異。最後，我們歸納出一套決定叢集網頁伺服器能力值的方法。

Because of the flourishing development of the internet applications, the services offered by popular web sites become diversified. Usually, the characteristics of these services are real-time and dynamic. Therefore, using only one single server to serve all the requests will always be overloaded. In this thesis, we adopt a flexible registration protocol to make the load balancing system easily to add or remove web servers according to the instantaneous situation. In addition, client requests can be distributed to the server by using Weighted Distributing Algorithm and users can obtain fair quality of service. We use a number of computers to simulate the large amount of client requests connecting to our load balancing system, and accessing the webpage randomly. The experimental results show that our system can distribute load according to the abilities of the servers for offering the dynamic webpage. We also compare the different abilities between single web server and our load balancing system. Finally, we will derive a method to get the ability of clustered web servers.

TABLE OF CONTENTS
中文摘要 i
ABSTRACT ii
TABLE OF CONTENTS iii
LIST OF FIGURES vi
LIST OF TABLES viii
CHAPTER 1 INTRODUCTION 1
1.1 Introduction 1
1.2 Motivation 1
1.3 Objective 3
1.4 Thesis Organization 3
CHAPTER 2 BACKGROUNDS 5
2.1 Server Architecture 5
2.1.1 High Performance Server 5
2.1.2 Server Cluster 6
2.1.3 Network Switch 6
2.1.4 Load Balance Equalizer 6
2.1.5 Program-oriented Mechanism 7
2.2 Server Cluster 7
2.2.1 Requirements to Build a Server Cluster 7
2.2.2 The Features of Server Clustering 9
2.2.3 The Architecture of Server Clustering 10
2.2.3.1 Centralized Connection Routing 10
2.2.3.2 Distributed Connection Routing 13
2.3 Distribution Technology 15
2.4 Connection Scheduling 19
2.4.1 Round-Robin Scheduling 19
2.4.2 Weighted Round-Robin Scheduling 20
2.4.3 Minimum Misses Scheduling 20
2.4.4 Hash Scheduling 21
2.4.5 Bandwidth Scheduling 22
2.4.6 Least-Connection Scheduling 22
2.4.7 Weighted Least-Connection Scheduling 23
2.4.8 Response Time Scheduling 24
2.5 Server Registration Protocol 24
CHAPTER 3 ARCHITECTURE AND IMPLEMENTATION 26
3.1 System Architecture 26
3.2 Flexible Registration Protocol 29
3.3 Capacity 30
3.4 Weighted Distributing Algorithm 31
CHAPTER 4 EXPERIMENTAL RESULTS AND ANALYSES 33
4.1 Hardware Platform 33
4.2 Experimental Operations 36
4.3 Analysis Results 36
CHAPTER 5 CONCLUSIONS AND FUTURE WORKS 48
5.1 Conclusions 48
5.2 Future Works 48
References 50


LIST OF FIGURES
Figure 2.1: Centralized connection routing architectures – TCP router with 2-way packet rewriting 11
Figure 2.2: Centralized connection routing architectures – TCP router with 1-way packet rewriting 12
Figure 2.3: Distributed connection routing architectures 14
Figure 2.4: DNS-based approach 16
Figure 2.5: TCP router approach 17
Figure 2.6: Network address translation approach 18
Figure 2.7: Load balancer and backend server message flows 25
Figure 3.1: The proposed system architecture 27
Figure 3.2: Server selection algorithm 28
Figure 3.3: Flexible registration protocol 30
Figure 3.4: Example of serverlist table 31
Figure 4.1: The steps of storing database on the ram disk 35
Figure 4.2: Maximum connection number per second for one to three web servers 37
Figure 4.3: Drop rate for one to three web servers 38
Figure 4.4: Average response time for one to three web servers 38
Figure 4.5: Connection number per second of three web servers with the same abilities 39
Figure 4.6: Maximum connection per second of web servers with different abilities 41
Figure 4.7: Drop rate of three web servers with different abilities 41
Figure 4.8: Average response time of three web servers with different abilities 42
Figure 4.9: Connections per second for capacity ratio of 9:11:13 44
Figure 4.10: Average response time for capacity ratio of 9:11:13 44
Figure 4.11: Connections per second under capacity ratio of 6:10:17 46
Figure 4.12: Average response time under capacity ratio of 6:10:17 46
Figure 4.13: Connection number ratio under capacity ratio of 6:10:17 47



LIST OF TABLES
Table 4.1: The experimental platform and parameters 33
Table 4.2: Serverlist table 40
Table 4.3: Comparison of three web servers 43
Table 4.4: Capacity of each web server under drop rate below 5% 43
Table 4.5: Capacity of each web server under drop rate below 5% and avg. response time below 1 second 45

References
[1]Zin-Chy Chi, Web Server Load Balance Architecture for Burst Mode Service Request, Master thesis, Department of Computer Science and Engineering, Tatung University, June 2004.
[2]Priya Kothari, “Clustering: A high availability solution,” the white paper of WIPRO Infotech, http://www.wipro.co.in/
[3]Chuen-Huo Wang, A distributed Linux cluster server system with fault-tolerant ability, Master thesis, Department of Electrical Engineering, National Cheng Kung University 2002.
[4]L. Aversa, A. Bestavros, “Load Balancing a Cluster of Web Servers Using Distributed Packet Rewriting,” in Proc. IEEE International Performance, Computing, and Communications Conference, pp. 24 - 29, February 2000.
[5]E. Anderson, D. Patterson, E. Brewer, “The Magicrouter: An Application of Fast Packet Interposing,” in Proc. Second Symposium on Operating Systems Design and Implementation, May 1996.
[6]Cisco Systems, “Scaling the Internet Web Servers,” the white paper of Cisco System, http://www.cisco.com/en/US/products/hw/contnetw/ps1894/products_white_paper09186a0080091edf.shtml
[7]IBM Corporation, “The IBM Interactive Network Dispatcher,” the redbook of IBM Corporation, http://www.redbooks.ibm.com/redbooks/pdfs/sg244993.pdf
[8]O. P. Damani, P. E. Chung, Y. Huang, C. Kintala, and Y.-M. Wang, “ONE-IP: Techniques for hosting a service on a cluster of machines,” Journal of Computer Networks and ISDN Systems, Vol. 29, No. 8-13, pp. 1019-1027, 1997.
[9]C. Yoshikawam, B. Chun, P. Eastham, A. Vahdat, T. Anderson, and D. Culler, “Using Smart Clients to Build Scalable Services”, in Proc. USENIX Technical Conference, Jan. 1997.
[10]DNS rfc, http://www.dns.net/dnsrd/rfc.
[11]Wensong Zhang, Shiyao Jin, Quanyuan Wu, “Scaling Internet Services by LinuxDirector,” in Proc. High Performance Computing in the Asia-Pacific Region, 2000.
[12]Tsang-Long Pao, Jian-Bo Chen, I-Ching Cheng, “An Analysis of Server Load Balance Algorithms for Server Switching,” in Proc. Ming-Chung University 2004 International Academic Conference, pp. 379-391, Mar. 2004.
[13]Wollman, W.V.; Jegers, H.; Loftus, M.; Wan, C., “Plug and Play Server Load Balancing and Global Server Load Balancing for Tactical Networks,” in Proc. Military Communications Conference (MILCOM), 2003, Vol. 2, Oct. 2003.
[14]Suntae Hwang, Naksoo Jung, “Dynamic Scheduling of Web Server Cluster,” in Proc. International Conference on Parallel and Distributed Systems (ICPADS), 2002.
[15]KNOPPIX, http://www.knoppix.net/