The most common type of storage for Exchange Server is direct-attached storage (DAS). DAS has existed in one form...
By submitting your email address, you agree to receive emails regarding relevant topic offers from TechTarget and its partners. You can withdraw your consent at any time. Contact TechTarget at 275 Grove Street, Newton, MA.
or another for decades, so I don't want to waste your time by talking about how to connect hard disks to your Exchange Server. Likewise, I want to stay away from talking about file placement and the various types of RAID arrays, because I covered those topics in Part 2.
Instead, I want to talk about how you can squeeze the best possible performance out of your DAS. When you are planning DAS for Exchange Server storage, there are two cardinal rules to keep in mind. The rules are simple, but they make a huge difference in your server's overall performance.
Rule 1: Two disks are better than one
A classic example: A friend of mine recently got a job as a network administrator for a different company. The company was still using Exchange 5.5, so he decided to upgrade them to Exchange 2003. The problem was that the server didn't have enough hard disk space for the upgrade. The existing server's RAID array contained six 40 GB hard drives.
My friend decided to replace those six drives with four 200 GB drives. Although the server's new configuration gave it a lot more disk space, he started getting calls from users because Exchange Server was responding slowly to their requests.
The problem: He had decreased the number of spindles available to Exchange Server, and therefore lowered the total data throughput.
Rule 2: Different RAID arrays have different amounts of overhead
In a typical Exchange Server deployment, the database has a ratio of somewhere between two and three read operations for every write operation (2:1 or 3:1). In an Exchange server without a RAID array or in an Exchange server with a RAID 1 array, this ratio is completely irrelevant because reads and writes have the same impact on disk I/O.
However, a RAID 5 array actually performs four different disk I/O operations for every write (two reads and two writes because parity information must be read, calculated, and written). Therefore, in a RAID 5 array, a write operation has quadruple the impact on disk I/O than a read operation. Similarly, a RAID 0+1 array performs two operations for every write, so writes have double the impact on disk I/O than reads.
Microsoft recommends that your hard disk not consistently work at more than 80% capacity. So, if you want to determine the maximum realistic throughput of a hard disk, you take the disk's physical I/O limit and multiply it by .8 (80%). For example, a drive that's rated at 130 I/Os per second would have a maximum realistic throughput of 104 I/Os per second.
But, if the disk is a part of an array (other than a RAID 1 array), this formula doesn't work. You then have to factor in the extra I/Os consumed by write operations. If you are assuming a three to one read/write ratio, you would use .45 as the multiplier for a RAID 5 array, and .64 for a RAID 0+1 or for a RAID 10 array.
Assuming that all things are equal, this means that you will get lower performance out of the DAS disks in a RAID 5 array than you would get from the disks in a RAID 0+1 array.
The chart below assumes a read/write ratio of three to one, and a physical disk limitation of 130 I/Os per second. I have rounded calculations to the nearest whole number.
|No RAID or RAID 1||RAID 5||RAID 0+1|
|Drive's physical capability||130 I/Os per second||130 I/Os per second||130 I/Os per second|
|Maximum realistic throughput||104 I/Os per second||59 I/Os per second||85 I/Os per second|
As you can see, the RAID arrays that dedicate I/O cycles to providing fault tolerance deliver fewer usable I/Os per second than arrays or individual disks without parity.
This is where Rule 1 comes back into play though. Two disks really are better than one. The calculations above are based on a single disk. The actual number of usable I/Os per second is derived by multiplying the calculated value by the total number of disks in the RAID array. Therefore, a single disk in a RAID 5 array is only delivering 59 I/Os per second. But, if there were four disks, the array as a whole could theoretically deliver about 236 I/Os per second.
CRASH COURSE: EXCHANGE SERVER 2003 STORAGE MANAGEMENT
Part 1: Microsoft's Exchange Server storage recommendations
Part 2: RAID configuration options for Exchange Server storage
Part 3: Using a SAN for Exchange Server storage
Part 4: Using NAS for Exchange Server storage
Part 5: Using DAS for Exchange Server storage
Part 6: Related resources on Exchange Server storage management
|ABOUT THE AUTHOR:|
| Brien M. Posey, MCSE
Brien M. Posey, MCSE, is a Microsoft Most Valuable Professional for his work with Exchange Server, and has previously received Microsoft's MVP award for Windows Server and Internet Information Server (IIS). Brien has served as CIO for a nationwide chain of hospitals and was once responsible for the Department of Information Management at Fort Knox. As a freelance technical writer, Brien has written for Microsoft, TechTarget, CNET, ZDNet, MSD2D, Relevant Technologies and other technology companies. You can visit Brien's personal Web site at http://www.brienposey.com.