Implement Of Event Injection System For Availability Evaluation Of Mission-critical Computer

Availability is one of the most important metric in computer system services evaluations,particularly in some special circumstances on some mission-critical computer, any failure of the system would be catastrophic. Hence, it's significant to ensure the availability. For the mission-critical computer, it's impossible to get the testing data through long time of waiting or observing system logs, as for these computers, the faults usually be small probability incidents. As a result, it's necessary to create failures artificially and collect system performance and system data. Fault injection technology is widely used in availability evaluation of mission-critical computer. However, existing single fault injection tools are unable to meet the high demand of mission-critical computer availability evaluations. Thus, a new fault injection model for mission-critical computer availability evaluations is proposed.The main work contains three parts: (1) Event injection model based on combination queuing method ,(2) Adjusted Impact Injection Method (AIIM) based on the local principle of program ,(3) Design and implement of event injection system for availability evaluation of mission-critical computer.Firstly, through analyzing the reason that existing fault injection tools which can't test mission-critical computer comprehensively. A new event injection model based on combination queuing method is proposed. On the basis of the switching control, this model can achieve single property test of the system, also can implement multi-angular measurement by the combined injection method. Timer introduced into the event injection tool can implement timing control for the injection system, so that we can achieve the aims of comprehensive evaluation for mission-critical computer.Secondly, after analyzing current existing memory address space injection method find out the issues of low success rates and high resource consumption. An Adjusted Impact Injection Method (AIIM) based on the local principle of program is proposed. By dynamically adjusting the impact times of a single experiment, it can guarantee events injection success rate and achieve the goal of saving resource at the same time.Finally, the event system is designed and implemented with modular design method,including registers event injection tool, memory event injection tool and I/O events injection tool. For those mentioned above, memory event injection tool contains memory overwrite event injection functional module, stack overflow event injection function module for parallel MPI program and memory leak event injection function module. I/O events injection tool contains read-write pointer offset event injection function module. With overall control module, the event injection system can build an automatic testing.