It Wasn’t My Fault: Understanding OS Fault Propagation Via Delta Execution

Recent approaches to operating system (OS) crash recov-ery have attempted to design a high-coverage component-agnostic recovery infrastructure [1, 3]. To successfully re-cover from an otherwise-fatal crash, it is necessary to deter-mine a safe execution point to resume operation consistently. Common restart strategies attempt to bring back the OS from a faulty execution state to a safe state and selectively re-play execution. Unfortunately, a faulty execution leading to a crash can result in many logical inconsistencies with complex dependencies among execution contexts (e.g. kernel threads) and OS subsystems that need to be tracked and rollbacked if necessary. Both heavyweight [3] and lightweight mecha-nisms [1] have been recently proposed for this purpose. Nev-ertheless, conclusions cannot be easily drawn on the effec-tiveness of these strategies, since very little is known abou