Rio File Cache: Surviving OS Crashes

Chen, et. al.

High-level idea: make volatile memory permanent. on an OS crash, do a warm reboot. on the warm reboot, dump all of physical memory to a swap file before initializing FS and VM sub-systems.

Advantages:
* Performance. Most systems assume memory is unreliable and have to come up with all kinds of crazy strategies to not write things to disk as often and batch writes (LFS, Group Commit, etc). With this strategy, all of these “reliability-induced” writes do not need to be done, since a regular write to memory is reliable.

Implementation:

How to achieve reliable, permanent memory?

need to protect memory... deactivate TLB write bits.

Protect buffer cache.