Synchronization method for loosely coupled arrays of redundant disk

Abstract

A multiprocessor system includes a plurality of substantially identical nodes interconnected through a switching network, each node including a disk drive, NVRAM, and a processor. The system stores data in either a RAID or mirrored fashion across a plurality of disk drives in different nodes. When data is stored in a RAID arrangement, an NVRAM in a parity node is provided with an entry including the new data, a copy of old data from the node to which the new data is to be written, a copy of the old parity, and a synchronization state indicator. The parity node determines new parity and transmits the new data to the data node for storage. Upon receiving an acknowledgement, the parity node resets the synchronization indicator. When power-up occurs, after a power failure, the parity node scans its NVRAM for any entry and upon finding one with a non-reset state indicator, transmits the new data to a destination data node for entry thereby synchronizing the contents of data and parity nodes. In a mirrored system, NVRAM in only one node has a data identifier entered into its NVRAM so that, upon a power failure and subsequent power-up, that entry enables the system to know which disk drives are in a non-synchronized state, and to cause actions that result in re-synchronization.