When it comes to providing disk protection for business data, one size does not fit all. Applications and even various categories of data within a given application require different levels of protection and performance. While RAID 1 might be suitable for one type of data, it might not be suitable for another. In part one of this tip, you'll learn about the individual types, or levels of RAID, and how much disk space is used to provide the relevant level of protection.
Essentially, RAID 0 does not protect data. Several disks are "striped" together to form a single amount of available disk space. Take five 100 GB disks, and after striping there is a single set of 500 GB that can be used for data. The failure of a single disk will require you to restore all of the data that is on the stripe.
RAID 0+1 uses a full second set of disks and mirrors the first set of disks. A single disk failure in one stripe does not cause any problems and the failed disk can be replaced and the missing information reconstructed. If you have an application that needs very high disk performance, but you don't have sufficient servers to allow one of them to go down in the event of a disk failure, then RAID 0+1 is an option for you.
This is probably the most common level of disk protection. Just about every conventional application server has two disks forming a mirror where the operating system is installed. The two disks must be identical in speed and size, and all data is written to both disks simultaneously by the RAID controller in the server.
Should one disk fail, the data is fully replicated to the other and no interruption to service is witnessed. Reads are taken from whichever of the two heads is closest to the data and all writes are simultaneously written to both disks in the pair. The pair appears to be only one disk to the operating system, and the hardware RAID controller takes care of managing the data.
This level of RAID is defined as a stripe with distributed parity. RAID 5 requires at least three disks, all of them identical in speed and size. All data is written across all of the disks except one and then the disk parity calculation is written to the remaining disk.
The next write is written, again, to all disks except one. But unlike RAID 4, the parity is not on a dedicated disk; each write progressively moves the parity to a different disk. RAID 5 can only cope with a single disk failure and when any disk fails, the data is reconstructed by the disk controller computing what data should be on the replacement disk.
RAID 4 and 6
RAID 6 takes an older standard, RAID 4, which uses a dedicated parity disk and adds a second parity disk to the array. However, this second parity disk is not necessarily a mirror or copy of the first parity disk. The parity information may be calculated in a different manner, using different algorithms, so that if two disks fail in the disks holding the data, the information can be reconstructed and efficiently delivered to the users.
Some vendors' implementations of RAID 6 involve far more disk traffic per write, which can degrade performance. Other vendors use intelligent caching and can lay a large write across the disks in a single operation, giving a high level of resiliency at the performance of RAID 10.
RAID 10 or RAID 1+0
Taking a number of RAID 1 pairs and forming a stripe out of them creates a RAID 10 array. If there are ten 100 GB disks arranged into five RAID 1 mirrors, the five pairs may be presented as a single entity by making a RAID 10 array. Thus the terabyte of raw disk turns into 500 GB of individual RAID 1 pairs. This allows the presentation of a single 500 GB array where previously some stitching at the file or partition level may have been necessary.
Other RAID levels
There are other RAID levels, noticeably filling in the gap between RAID 1 and RAID 5. In the early days of server computing, these levels served a particular purpose. However, they have become uncommon as disk capacities and reliability have grown, and as computing and storage has become more commoditized. The levels mentioned above form the bulk of disk protection capabilities in the market today.
About the author: Mark Arnold, MCSE+M, Microsoft MVP, is the principal consultant with LMA Consulting LLC, a Philadelphia, PA-based private messaging and storage consultancy. Mark assists customers in designs of SAN-based Exchange implementations. You can contact him at [email protected]
Do you have comments on this tip? Let us know. Please let others know how useful this tip was via the rating scale below.
Do you know a helpful storage tip, timesaver or workaround? Email the editors to talk about writing for SearchSMBStorage.com.
The benefits of RAID and where it works best