Xfas: Fault Recovery by Externally Controlling OS Behavior
Kento Kimura, Kenichi Kourai · 2023
It is difficult to completely avoid system failures in recent large-scale and complex clouds. Therefore, it is important to detect system faults as symptoms of failures and then recover from them. Admins often attempt fault recovery by remotely logging in to the target system. They can also run recovery systems inside the target system in advance. However, both methods are subject to system faults in the target system. In this paper, we propose Xfas for recovering from system faults by indirectly controlling OS behavior from the outside of the target system. Xfas attempts fault recovery by rewriting OS data in the memory of the target system and leveraging the capabilities of the OS itself. For example, it mimics signal sending to force termination of abnormal processes and unlocking to recover from some kind of deadlock. As two instances of Xfas, this paper presents VMMfas and GPUfas. We confirmed the effectiveness and efficiency of fault recovery by Xfas.