Hardware performance counters for system reliability monitoring

As technology scaling reaches nanometre scales,the error rate due to variations in temperature and voltage,single event effects and component degradation increases, makingcomponents less reliable. In order to ensure a system continuesto function correctly while facing known reliability issues, it isimperative that the system should have the means to detect theoccurrence of errors due to the presence of faults. A system thatbehaves normally (no error detected in the system) exhibits aprofile, and any deviations from this profile indicate that thereis an anomaly in the system. In this paper, we propose to usehardware performance counters (HPCs) to measure events thatoccur during the execution of the program. We explore thevarious counters available which could be use to identify theanomalous behaviour in the system and develop a methodologyto observe the anomalies using HPCs by creating a faultfreepattern and observing any subsequent changes in thatpattern. We evaluate the proposed technique using GemFI, anarchitectural simulator based on Gem5 with additional faultinjection capabilities. We compare the results obtained at theend of the execution with data collected during a time interval.Our results show that HPCs can be used to identify anomalousbehaviour in a system that would lead to failure