隨著全球資訊網網路環境的普及，以及具有平台獨立特性的爪哇位元碼，現在數以百萬計具有爪哇執行能力的電腦可以互相連線，用來分享它們的計算能力。安裝在許多不同電腦上的爪哇虛擬機器，被整合成一個計算池，取代一部昂貴的大型電腦來執行計算密集的應用程式。為了提供一個穩健的分散式計算環境，本文提出“以爪哇為基礎之分散式計算系統上的容錯架構”。本架構提供的最主要優點是我們提供一種可隨著網路狀況而做調整的容錯架構，並且可用於原來單機的爪哇程式碼。像公共計算池這樣不穩定的網路環境，遠端方法呼叫仍然缺乏一個穩健的機制，可以確保每個運算在一次派遣中完成。本架構的應用程式介面延伸遠端方法呼叫的應用程式介面並結合主動複製機制，程式設計師只要延伸本架構的應用程式介面，不需要修改其程式內容即可做穩健的分散式處理。一般來說，一個應用程式是由許多運算合作完成。本架構複製每個運算，並同時分派它們到不同的運算點；當計算點之一完成該複製品的運算時，此運算便告完成。在非常不穩定的網路狀況下，我們也可以增加複製的數量，來確保至少有一個運算點能完成該運算。

With advances in widespread networking, public WWW environment, and platform-independent Java bytecode, millions of Java-capable computers can be connected for sharing computing ability now. These heterogeneous supercomputers, workstations, personal computers, and laptops, can be merged as a pool of distributed Java virtual machines and exploit their large number of computing cycles for CPU-intensive applications. In order to provide a robust distributed environment, a Fault-Tolerance Framework for Java-Based Distributed Computing System (FJDCS) has been proposed in this thesis. The most important advantage of our system is providing an enhanced and configurable fault-tolerance mechanism to all of legacy Java applications. In the very unreliable networking environment like public computing pool, the RMI mechanism still lacks a robust fault-tolerance mechanism to ensure that every computation can be completed in an iteration. We extended the RMI API and combined the replication mechanism that can be categorized to active replication mechanism to build our FJDCS API. Programmers can just extend our API directly and do not need to modify their legacy applications to get our robust fault-tolerance mechanism. In most cases, an application is completed by many cooperated tasks. In the proposed system, we replicate every task by two or more instances and dispatch them to the different computing nodes concurrently. When one of the computing nodes that process the instances of the same task has completed its operation, this task is completed. In the very unreliable network, we can configure the number of clones for one task to ensure that at least one computing node can complete this task.

Abstract in Chinesev
Abstractvi
Contentsviii
List of Figures & Tablesx
Chapter 1 Introduction1
1.1 Motivation1
1.2 System Design2
1.3 Organization of This Thesis3
Chapter 2 Background4
2.1 Fault-Tolerance Mechanisms4
2.2 Remote Method Invocation5
2.3 Object Serialization9
2.4 Security Manager9
2.5 Applets10
2.6 CORBA11
2.7 Related Works11
Chapter 3 Fault-Tolerance Framework For Java-Based Distributed
Computing System14
3.1 System Overview14
3.2 Implementation17
3.2.1 System Hierarchy17
3.2.2 Implementation of the System Elements18
Chapter 4 Experiments22
4.1 The Assumption of the System22
4.1.1 Overview22
4.1.2 Distribution of the Assumed Queueing Model24
4.2 Performance of FJDCS24
4.2.1 smpl Simulation Subsystem25
4.2.2 Simulation Model27
4.2.3 Simulation Results28
4.3 Data Dependency Situation30
Chapter 5 Conclusion and Future Work32
5.1 Conclusion32
5.2 Future Work32
References33

[1]SETI@home. http://setiathome.ssl.berkeley.edu/, 2001
[2]Sun Microsystems, Inc. Java Remote Method Invocation — Distributed Computing For Java. http://java.sun.com/marketing/collateral/javarmi.html, 2001
[3]Sun Microsystems, Inc. Object Serialization.
http://java.sun.com/j2se/1.3/docs/guide/serialization/, 2001
[4]Navin Budhiraja, Keith Marzullo, Fred B. Schneider, and Sam Toueg. The Primary-Backup Approach. Addison-Wesley Publishing Company, second edition, 1993
[5]Rachid Guerraoui and Andre Schiper. “Software-Based Replication for Fault Tolerance,” IEEE Software, May 1995. pp. 29-41
[6]David Powell and Paulo Verissimo. “Delta-4: A Generic Architecture for Dependable Distributed Computing,” Springer-Verlag, 1991
[7]A. J. Wellings and A. Burns. “Programming replicated systems in ada 95,” The Computer Journal, 39(5), 1996. pp. 361-373
[8]David Flanagan. JAVA FOUNDATION CLASSES, ISBN: 1-56592-488-6. O’Reilly & Associates, Inc., 1999
[9]Sun Microsystems, Inc. CORBA Technology and the Java Platform. http://java.sun.com/j2ee/corba/
[10]Jim Farley. JAVA Distributed Computing, ISBN: 1-56592-206-9. O’Reilly & Associates, Inc., 1998
[11]G. A. Geist, V. S. Sunderam. “The PVM system: supercomputer level concurrent computation on a heterogeneous network of workstations,” Proceedings of the Sixth Distributed Memory Computing Conference, IEEE, 1991. pp. 258-261
[12]David A. Thurman. “jPVM - An interface written using the Java native methods capability.” http://www.chmsr.gatech.edu/jPVM/, 1998
[13]Adam J. Ferrari. “JPVM — The Java Parallel Virtual Machine.”
http://www.cs.virginia.edu/~ajf2j/jpvm/, 1999
[14]M. Snir, S.W. Otto, S. Huss-Lederman, D.W. Walker, J. Dongarra. MPI, The Complete Reference. MIT Press, Cambridge, November 1995
[15]Gilbert H. Young, Lai-man Wan and Vincent S. Yiu. “JAVA MESSAGE PASSING INTERFACE,” Proceedings of IEEE 1997 National, Vol. 1, 1997. pp. 189-194
[16]Kivanc Dincer. “A Ubiquitous Message Passing Interface Implementation in Java: jmpi,” Parallel and Distributed Processing, 1999. 13th International and 10th Symposium on Parallel and Distributed Processing, 1999. 1999 IPPS/SPDP. Proceedings, 1999. pp. 203 —207
[17]M.H. MacDougall. Simulating Computer System, Techniques and Yoold. MIT Press Series 1989.