由於網際網路的盛行，為了應付熱門網站龐大的網路連線，採用多台計算機構成之叢集伺服器系統來分擔負載是一可行的辦法。而在分散式叢集伺服器系統設計上，由美國波士頓大學Azer Bestavros所提出之分散式封包重寫(DPR, Distributed Packet Rewriting)技術則被廣泛加以討論著。此分散式叢集伺服器系統中會採用領域名稱服務輪詢系統(RR-DNS, Round-Robin Domain Name System)來實現，但由於領域名稱服務輪詢系統是種階層式架構，且為了增進效率而有著快取的行為，所以較容易導致系統上動態負載失調；此外當某台伺服器因故障而當機候，遠端客戶端假如果還是利用原先所快取住的舊資訊來連線的話，可能就會導致錯誤連線的結果。因此，如何改善這些問題是一個值得研究的課題。
　　在分散式系統中利用領域名稱服務輪詢系統所產生的負載不平衡，可利用負載平衡演算法來加以改善。而對於產生之連線錯誤則需錯誤容忍之機制來克服。在本論文中我們提出了一個鏈狀夥伴(chain- buddy)之容錯法：此容錯法可讓在同一子網域的每台伺服器各自去監視另外一台伺服器的狀態，而形成一個鏈狀監管拓樸架構。當系統中某台伺服器無法提供服務的時候，負責接管它的伺服器會馬上去接收它的IP位址及提供原先的服務，而一旦故障的伺服器恢復時，再歸還原先的IP位址，以達到容錯效果。
　　本論文之系統採取httperf測試軟體，加以測試單機伺服器和叢集伺服器兩者在效能和動態負載不平衡改善的狀況，及測試鏈狀夥伴容錯方法是否有容錯的效果。實驗結果證明分散式封包重寫狀況加上負載平衡演算法對於領域名稱服務輪詢系統之負載不平衡之缺點能加以改善，而本論文所提出之的鏈狀夥伴容錯法也充分發揮了容錯的效果。

　　With the explosive growth of the Internet, some popular web sites use cluster server for tens of thousands of hits per second. To build high performance Web Servers, designers tends to distributed systems. The DPR(Distributed Packet Rewriting)system proposed by Azer Bestavros raises a great discussion. The DPR system uses RR-DNS(Round-Robin Domain Name System) for distribute requests , but brings some problems – dynamic load imbalance and probable connection error.
　　To solve the issue of connection load imbalance, we can use appropriate load balanc-
ing algorithm for every incoming packets. To avoid the probable connection error caused by RR-DNS, our essay proposes a chain-buddy way with fault-tolerant ability：every server in the same subnet monitors its buddy to check if it is dead or not and the circumstance forms the chain topology. When there is a broken-down server occurred in this system, the one who takes over it gets its IP address and continues the original service which it supports. Once the broken-down server recovers, the one who takes over it return its IP address and the system achieves the fault-tolerant effect.
　　To verify the correctness, we use the simulation benchmark software – httperf which develops by HP company. The experiment shows that the DPR system with load balancing algorithm and chain-buddy way has the feature of high performance and can remove the shortcomings derived by RR-DNS , that it becomes a distributed fault-tolerant system

[1] D.Andresen et al., ”SWEB: Towards a Scalable World Wide Web Server on Multicomputers ,”Proceedings of the 10th International Parallel Processing Symposium , Honolulu,Hawaii, pp. 850-856(April 1996)

[2] L.Aversa and A.Bestavros, ”Load Balancing a Cluster of Web Servers Using Distributed Packet Rewriting,” Proceedings of the 2000 nternational Performance, Computing, and Communication Conference, Phoenix ,Arizona, pp. 2-29(February2000)

[3]A.Bestavros et al., “Distributed Packet Rewriting and its Application to Scalable Server Architectures,“ Proceedings of the international Conference on Network Protocols, Austin Texas, pp.290-297(October 1998)

[4]Cisco System Inc. “Failover Configuration for LocalDirector,” Cisco System White Paper– http://www.cisco.com/warp/public/cc/pd/cxsr/400/tech/locdf_wp.htm

[5] Aghdaie, N.; Tamir, Y. “Client-transparent fault-tolerant Web service ,“Performance, Computing, and Communications, 2001. IEEE International Conference on. , 2001 Page(s): 209 –216

[6] Wensong Zhang; Shiyao Jin; Quanyuan Wu ,“Scaling Internet services by LinuxDirector, “High Performance Computing in the Asia-Pacific Region, 2000. Proceedings. The Fourth International Conference/Exhibition on , Volume: 1 , 2000 Page(s): 176 -183 vol.1

[7] Chu-Sing Yang; Mon-Yen Luo , “Building an adaptable, fault tolerant, and highly manageable web server on clusters of non-dedicated workstations,” Parallel Processing, 2000. Proceedings. 2000 International Conference on , 2000 Page(s): 413 –420

[8]O.Daman, P.Chung, Y.Huang,C.Kintala,and Y.Wang ,”ONE-IP: Techniques for Hosting a Service on Cluster of Machines,” Computer Network and ISDN System,29,1997

[9]E.Anderson ,D. Patterson,and E.Brewer, “ The Magicrouter ,an application of fast packet interposing,”http://www.cs.berkeley.edu/~eanders/projects/magicrouter/osdi96-mr-submission.ps

[10]C.YoSHIKAWA, B chun,P.Eastham,A.vahdat,T.Anderson,and D. Culler,”Using smart clinets to build scalable services,” Proceedings of the 1997 USENIX Annual Technical Conference,pp. 105-117, January 6-10 ,1997

[11] T. Brisco “DNS Support for Load Balancing”, Network Working Group ,RFC1794 http://www.faqs.org/rfcs/rfc1794.html

[12] C.Perkins ,”IETF RFC2003：IP Encapsulation within IP”, Available from ftp://turtle.ee.ncku.edu.tw/pub/documents/rfc/rfc2003.txt

[13] Dias, D.M.; Kish, W.; Mukherjee, R.; Tewari, R.”A scalable and Highly Available
Web Server” Compcon '96. 'Technologies for the Information Superhighway' Digest of Papers , 1996 Page(s): 85 -92

[14] Rusty Russell, “Linux 2.4 Packet Filtering HOWTO” , available from http://www.linux.org.tw/CLDP/Packet-Filtering-HOWTO.html,2000

[15]G.Wright and W.R.Stevens ,TCP/IP Illustrated,Volume 1, Addison-Weley,Reading, May 1994

[16] Mon project：http://www.kernel.org/software/mon/

[17] fake Project：http://www.vergenet.net/linux/fake/

[18] HeartBeat project：http://www.linux-ha.org/

[19] David Mosberger , Tai Jin,”Httperf-A Tool for
Measuring Web serve performance ”, HP Research Labs,Available from http://www.hpl.hp.com/personal/David_Mosberger/

[20] http://www.linux-vs.org/HighAvailability.html

[21]Linux Virtual Server Projects ：http://www.linux-vs.org/

[22]IBM Corporation. The IBM Interactive Network Dispatcher. See http://www7.scu.edu.au/programme/fullpapers/1899/com1899.htm

[23] T. Berners-Lee, R. Fielding and H. Frystyk, “Hypertext Transfer Protocol –1.0”
http://www1.ics.uci.edu/pub/ietf/http/rfc1945.html

[24] P. Mockapetris.” DOMAIN NAMES - CONCEPTS AND FACILITIES.
ftp://ftp.ee.ncku.edu.tw/pub/documents/rfc/rfc1035.txt

[25] Cisco Systems. “Scaling the Internet Web Servers”. A white paper available from
http://www.cisco.com/warp/public/cc/pd/cxsr/400/tech/scale_wp.htm

[26] http://www.apache.org/

[27] http://www.isc.org/products/BIND/

[28] 連結共享儲存設備http://www.linux.org.tw/CLDP/High-Availability-HOWTO/High-Availability-HOWTO-7.html#chg-ultra-scsi

[29] ASPAC計畫, “Gnuplot導讀”，中央研究院計算中心，available from ftp: ftp://phi.sinica.edu.tw/pub/aspac/doc/95/95006.ps

[30] 許順興,何裕琨,”利用最少工作量與連線數演算法作動態負載平衡之Linux叢集伺服器“ national computer Symposium,2001Page(I212-I220)