Instant OS updates via userspace checkpoint-and-restart
Sanidhya Kashyap, Changwoo Min, Byoungyoung Lee, Taesoo Kim, Pavel G. Emelyanov · USENIX Annual Technical Conference · 2016
In recent years, operating systems have become increasingly complex and thus more prone to security and performance issues. Accordingly, system updates to address these issues have become more frequently available and increasingly important. To complete such updates, users must reboot their systems, resulting in unavoidable downtime and further loss of the states of running applications. We present KUP, a practical OS update mechanism that employs a userspace checkpoint-and-restart mechanism, which uses an optimized data structure for checkpointing on disk as well as a memory persistence mechanism across the update, coupled with a fast in-place kernel switch. This allows for instant kernel updates spanning across major kernel versions without any kernel modifications. Our evaluation shows that KUP can support any type of real kernel patches (e.g., security, minor or even major releases) with large-scale applications that include memcached, mysql, or in the middle of the Linux kernel compilation, unlike well-known dynamic hot-patching techniques (e.g., ksplice). Not only that, KUP can update a running Linux kernel in 3 seconds (overall downtime) without losing 32 GB of memcached data from kernel version v3.17-rc7 to v4.1.