High reliability, high performance disk array storage system

摘要

A system for ensuring high reliability in a block service disk array system while promoting high performance by logically writing all changes to strides on the array while physically writing ahead to a log only a subset of the changes. Specifically, for changes of only a strip or so, the changes are written to a log, along with a commit record, and then written to disk, later deleting the changes from the log. In contrast, for relatively larger changes, i.e., for an entire (or nearly entire) stride, the old stride is not overwritten by the new, but rather is written to a new location on the disk, with the new and old locations and a commit record (but not the new stride itself) being logged and with the entries for the locations in the stride mapping table swapped with each other. In an alternate embodiment, blocks can be written to temporary locations in a RAID-1 area and lazily moved to home locations in a RAID-5 area of an array of disks.