The Design and Realization of Evaluation Platform Based on Faults Injection in Massive Storage

摘要

A platform including a set of evaluation tools is designed in this paper aimed at the research of fault tolerance and availability in multi-levels Massive Storage. The platform implements faults injection on network drive level and distribution file system level. Then the evaluation results can be obtained from the analysis results after fault injected. On the network drive level, I/O faults pattern in SCSI disk are defined. SCSI instructions are modified in order to simulate SCSI disk faults by intercepting functions in SCSI medium and bottom layer. On distribution file system level, faults assembly involves most faults in server layer, catalog layer and file layer. The test targets system is modified including service continuity, outward service latency time, and performance demotion ratio from which systems can be compared. The platform has been used for our project testing. From the testing results, different fault tolerance and availability can be obtained in different storage system effectively.