RAID: How safe is your data?
By Brett Cooper
RAID, in its current designs, can handle a single-disk failure gracefully. However, in the case of a double-disk failure in a given RAID volume the results can be disastrous. There is only one solution, that I am aware of, that can withstand a double-disk failure in the same RAID group without having a full mirror of the RAID volume. It is the Network Appliance RAID Diagonal Parity (RAID DP), in which the parity bits for RAID are striped both horizontally as with most standard RAID solutions, but also diagonally, thus protecting the environment against a double-disk failure. This is done with only a single extra parity disk per RAID group and, as a result, offers excellent cost/feature benefits. For more information on RAID DP, please visit www.netapp.com/tech_library/3298.html
For other errors, such as power outages, different solutions are required. In the case of a power outage, there a variety of products, both in the disk subsystem and outside the system, such as redundant power supplies on separate circuits that are conditioned and backed up by batteries (UPS) or generators.
Many companies are using battery backed up NVRAM (nonvolatile random access memory) that stores a log of the information being written from the host prior to it being committed to disk. This log will recover a storage system up to the point that the power went out and will guarantee that data isn't lost. As far as data loss is concerned, there are various areas of protection against this, the simplest is a real-time snapshot of the information on the disk subsystem, followed by a full tape backup and then by a mirror of the information. In all of the above cases, consideration must be made and the trade-offs weighed in terms of complexity, risk and cost.
If you need help laying out a solution for a specific application, it is best to check with the vendor of that application for some best practices on designing the solution and then engage the rest of the vendors involved in the solution. You may even need to bring in a services or consulting organization to help you design the best environment for your application.
01 Jun 2005