Improving RAIDframe parity handling

by Jed Davis for The NetBSD Project

If a NetBSD system using a software RAID device loses power or crashes, a lengthy (many hours, for today's disk sizes) and I/O-intensive parity rewrite is required afterwards. I propose to modify the raid(4) driver to remove this deficiency — which may be discouraging the use of NetBSD in server applications — by more closely tracking which parity information actually needs to be recomputed. In addition, some less major shortcomings of raid(4) could be addressed at the same time.