Algorithms for the management of remote backup data bases for disaster recovery

Abstract

For high availability in disaster recovery situations, it is desirable to reflect data base changes made by a transaction processing system continuously on a (replicated) data base maintained at a remote site. To keep the resource consumption on the backup system low, we describe a general, low overhead method for exploiting parallelism in the processing of the log records received at the backup system. Our approach is general enough to accommodate even the ARIES-type recovery and concurrency control methods which support high concurrency and high efficiency via write-ahead logging, nested transactions, operation logging and semantically-rich modes of locking. We also propose techniques for (1) checkpointing the state of the backup system so that recovery can be performed quickly in case the backup system fails, and (2) allowing new transaction activity to begin even as the backup is taking over after a primary failure. We present some performance measurements from a prototype. We discuss distributed transactions, shared disks (data sharing) transaction environment and combining executions of 1-Safe and 2-Safe transactions in a single system.