Snapshots and Continuous Data Replication in Cluster Storage Environments

Abstract

Snapshots are an elegant means to derive instant, virtual copies of storage devices. For copy-on-write snapshots, data has only to be physically copied, if either the content of the original device or the content of the snapshot device changes. The online process of copying data between the devices can have a severe impact on application performance, especially if more than a single snapshot is derived from the original volume. This performance decrease becomes even worse, if the storage device is composed from individual storage bricks, working together as peer-to-peer storage cluster. This paper presents the architecture, the used data structures, and corresponding performance results for a snapshot implementation that is able to support an arbitrary number of snapshots from an original volume in a storage cluster environment, enabling it to perform (near) continuous data protection.