Redundancy is a common approach to cope with transient hardware errors
which activate bit flips in processors. The replication framework
Romain on top of L4/Fiasco, implements a redundancy mechanism that
detects and recovers from bit flips by using a majority decision of
correct and valid replicas. Therefore, Romain checks the replicas
whenever they encounter an exception/pagefault or issue a system call.
The goal of my work is to achieve configurable, frequent checks
of replicas in order to reduce the error detection latency with the help
of an instruction-based watchdog.
Bounding Error Detection Latencies for Replicated Execution
Verteidigung der Bakkalaureatsarbeit