The Design And Implementation Of Fault-Injection Platform For High-Performance Fault-Tolerant Computer Oriented Itanium Architecture

High-performance fault-tolerant computer systems are widely used in key areas such as banking, telecommunications, etc. Different from the ordinary transaction processing systems, high-end fault-tolerant computer systems must have very strong fault-tolerant mechanisms, high system availability and strong system process capacity. Otherwise, the system failure may cause delays and devastating disaster. Therefore, how to evaluate the availability of high-end fault-tolerant computer is very important and difficult. As an effective experimental method for the availability assessment, fault injection could accelerate system failure by artificially introducing faults into the system, and provide abundant data to support availability assessment through the collection and analysis of the system response to the failure.This paper firstly gives a survey on existing tools for evaluating the availability both at home and abroad. There are two common problems in the traditional tools:1) goal-oriented system is limited to IA-32 architecture and research on injecting faults into the 64-bit architecture system is scarce; 2) It only has a single fault-injection method and doesn't support adding new fault-injection means. In response to these problems, this paper focuses on the target system of Itanium architecture, designs and implements a platform through message flow controlling the process of fault injection. The platform puts a variety of dispersive tools for the Itanium architecture into a framework, and provides the united test interface to the testers to inject different levels of faults into the target system using various methods. In addition, this paper proposes a way to develop the fault-injection tools based on components, and designs a generic component design standard that assists the user when adapting the platform to new fault injection tools.The testing environment of the fault injection platform needs to cover a wide range of platforms, which increases the complexity of building environment. Generating test cases manually also extends the time of test. Considering the manual testing problems, this paper establishes a fault-injection test automation system based on Software Testing Automation Framework (STAF). By using process services and file system services, three modules (dynamic deployment, test task control, test monitor) are implemented, and a distributed fault-injection environment for test automation is built. Finally, for the HP RX6600 and Superdome, the effectiveness of developed platform has been verified by using the implemented fault-injection platform and test automation system to do the manual and automation tests.