As we work with development teams on building out data grid architectures, a common requirement has surfaced, which is to provide backup to a remote data center, supporting Disaster Recovery (DR) and High Availability (HA) for the complete application. With more than one source of data, consistency and synchronization between the data centers are important issues — but even more important is the requirement to avoid data loss during replication.
Recently, we’ve been under development with our new CloudTran Replicator Service. This addresses the requirement above using an active-passive configuration: one data center is primary, the other is backup. It may be possible to organize the application into separate domains (where transactions only involve data from one domain), which allows an active-active configuration: one data center is primary for half of the data set and backup for the other half.
Our approach makes transactions on the order of 10 to 100 times faster than typical “data center backup” approaches that reconcile transactional inconsistencies using a distributed voting procedure. This clearly impacts the response time to the user. Less obvious is impact on server capacity: reducing the length of time for a transaction reduces the amount of information that must be tracked in the server.
Ultimately, the CloudTran approach to data center replication results is less cost to buy and run servers… a great thing for most enterprises these days. For those with more than a casual interest, many of the technical details of our replicator approach can be found in this newly published white paper.