Ln= _Database Recovery Using Redundant Disk Arrays*

for public release and sale is Sv D distribution is unlimited U Abstract Another type of failure is media failure. One com-mon way to deal with this type of failure is by periodi-Redundant disk arrays provide a way for achieving cally generating archive copies of the database and by rapid recovery from media failures with a relatively low logging updates to the database performed by commit-storage cost for large scale database systems requiring ted transactions between archive copies into a redo log high availability. In this paper we propose a method for file. When a media failure occurs the database is re-using redundant disk arrays to support rapid recovery constructed from the last copy and the log file is used from system crashes and transaction aborts in addi- to apply all updates performed by transactions that tion to their role in providing media failure recovery, committed after the last copy was generated. In such A twin page scheme is used to store the parity infor- a case, a media failure causes significant down time mation in the array so that the time for transaction and the overhead for recovery is quite high. For large commit processing is not degraded. Using an analyti- systems, e.g., with over 50 disks, the mean time to cal model, we show that the proposed method achieves failure (MTTF) of the permanent storage subsyste