Faster updating of data on parallel disks.

摘要

A method to increase the rapidity with which a host computer can perform random write operations to a controller managing one or more disk storage devices. Blocks of data are known to the host by a fixed logical address, but are stored at a physical location on disk chosen at the discretion of the controller. Blocks of data written by the host are stored in buffers in the controller until a chosen number have been accumulated. The controller then selects a track on each of one or more disk storage devices and writes the buffered blocks to these tracks such that each track is written in one rotation of the disk and all disk storage devices are written to in parallel. The chosen number of blocks is KN, where K is the number of blocks in each track, and N is the number of disk storage devices. The controller maintains an indirect map to allow a block's physical location on disk to be determined from its fixed logical address. The indirect map is nonvolatile to prevent the loss of data in the event of a failure. An allocation map indicates which physical blocks are in use. Whenever the number of empty tracks that can be selected for writing drops below a threshold, the controller creates new empty tracks by transferring blocks to other locations. <IMAGE>