High availability for parallel computers

Abstract

Fault tolerance has become an important issue for parallel applications in the last few years. The parallel systems' users want them to be reliable considering two main dimensions, availability and data consistency. Availability can be provided with solutions such as RADIC, a fault tolerant architecture with different protection levels, offering high availability with transparency, decentralization, flexibility and scalability for message-passing systems. Transient faults may cause an application running in a computer system to be removed from execution, however the biggest risk of transient faults is to provoke undetected data corruption that changes the final result of the application without anyone knowing. To evaluate the effects of transient faults in the robustness of applications and validate new fault detection mechanism and strategies, we have developed a full-system simulation fault injection environment

Keywords

• fault tolerance
• availability
• radic
• transient faults
• performability