Ups, downs of continuous data protection.
Continuous data protection: great for data corruption, less effective for disaster recovery.
While there are various ways to implement continuous data protection, most experts agree that the defining characteristic of CDP is the ability to roll back a system, database or set of files to any time within the retention window. Generally, the detail that can be achieved with CDP depends on the frequency of file-system transactions, which can occur many times per second.
CDP has gained widespread support helping enterprises recover from data corruption. Historically, corruption recovery relied on slow processes with much less “time granularity” to recover, for example, from a tape backup or periodic database dumps. CDP provides much faster recovery and much greater restore detail. However, it does not generally help companies with disaster recovery, and this is an important distinction.
CDP systems typically have agent software that resides on the storage platform and/or server being protected and a dedicated application component on a separate machine. The agent is responsible for capturing and transmitting every file-system transaction (that is, every modification to files on the system) to the dedicated CDP application system. The CDP system then records each transaction and the precise sequence and timing of the transaction.
Most CDP systems keep a rolling log of the transactions on the systems they are protecting and do not maintain a complete copy of the data being protected. CDP systems typically offer a configurable time window for rollback, usually measured in hours or days. When transactions age beyond this window, they are discarded. Because CDP stores the transactions, not the complete file system, CDP is useful only for recovering from corruption, not for recovering from system or hardware failure.
Because CDP systems are capturing transactions below the level of the file system in an indiscriminate fashion, little system overhead is added. It is similar to having a local disk mirror, which is a mature and efficient process.
At the time of corruption, an administrator is able to look through a transaction history and pick a rollback time. Some technologies let the administrator roll the production system back to that time or spin off a virtual instance of the system as it existed at the rollback point and mount that version. By doing this, the administrator can assess whether the corruption issue is resolved before presenting that version to the production environment. Then, the administrator can accept that version and replace the production version or try again for a different time.
Because CDP tracks file-system changes, restoration or rollback is exceptionally fast, a matter of reversing disk transactions. It isn’t limited by file-system bottlenecks, such as file creations or deletions. This high performance makes CDP especially useful and appropriate for very large and critical databases. CDP systems usually take advantage of high-performance storage systems, such as storage-area networks.
CDP offers other attractive features. For example, it lets database administrators mount historic versions of production databases from any time within the history window of the production servers parallel to the production systems, with little production impact. This lets administrators do testing, run reports or perform forensics. Also, CDP can be used much like snapshots to perform backups or copy large files, such as databases, at an exact time without having to stop the database.
CDP has its limitations. It is generally not cost effective outside large businesses, and it doesn’t address system or disaster-recovery requirements. In most cases, it is an entirely complementary solution – it doesn’t offset any existing costs.
However, for companies with large, critical databases, CDP is an excellent complement to existing data-protection solutions, helping companies increase the availability of their most critical applications.
Nesnow drives growth strategies for Iron Mountain’s Data Protection businesses. He was chief technology evangelist at LiveVault before Iron Mountain acquired it. He can be reached at Geoffrey.nesnow@ironmountain.com.
