What Hard Drives Should I Use When Building a New Server?

Today I received an email from one of the LearnItFirst.com users, Donny, who had recently viewed my Windows Server 2008 training videos and had a follow-up question.

After watching the Windows 2008 video course, I have learned that Grant suggests 2 hard drives for a basic server, 1 for the os and 1 for storage. Have not looked at the Exchange videos yet but does that mean I need 3 hard drives for the server? 1 for os, 1 for storage, 1 for exchange? Why not different partions on the hard drives? Smallest hard drive I can get on a sata is 160 gig, so does that mean 2 160 gigs for mirror os and 1 hard drive for storage And do you recommend ide, sata, scsi, sas? Just would like a little advice from an expert about the right direction to go. With Novell, I use Seagate Cheetta SCSI hard drives with adaptec scsi controller.

With any server - including exchange - it's all about the spindles, not partitions.

An ideal entry level File server would be the following 2x Hard Drives - in a RAID1 Configuration (MIRROR) - so all data is on both drives ONE Partition - two partitions will slow things down - as it must jump back and forth between the partitions

An ideal mid-range file server would be the following 2x Drives for the C:\ in a RAID1 (Mirror) for holding OS and binaries 2x Drives for the E:\ in a RAID1 (Mirror) for holding your data (if you can do this in the hardware or on a RAID controller - performance hit is negligible)

An ideal High-End Exchange Server (Thousands of Users) 2x Drives for the C:\ in a RAID1 (Mirror) for holding OS and Binaries 2x Drives for the E:\ in a RAID1 (Mirror) for holding the transaction logs 3x or more Drives for the F:\ in a RAID5 (Striped Set with Parity) for holding the Exchange Database(s)

But - very few people need the high-end configuration, so we compromise An ideal Mid-Range Exchange Server (Hundreds of Users) 2x Drives for the C:\ in a RAID1 (Mirror) for holding OS, Binaries, and Transaction Logs 3x Drives for the E:\ in a RAID5 (Striped Set with Parity) for holding the Exchange Databases
An ideal Small-Range Exchange Server (Dozens of Users) (my MINIMUM CONFIGURATION) 2x Drives for the C:\ in a RAID1 (Mirror) for holding OS, Binaries, and Transaction Logs 2x Drives for the E:\ in a RAID1 (Mirror) for holding the Exchange Databases

RULE #1 of Exchange (or any other transaction-based system) - NEVER Store your Transaction Logs with your Databases. In order for a recovery - you need one of the following A current Database (so loss of the C:\ drive is OK - if you have Database on E:\) or A backup of the Database (so last night's backup and loss of E:\), plus the Current Transaction logs
    Recovery: Restore database from last night, database will use the logs to return to normal operation)

What about the partition idea and why split everything?
A partition logically divides the disk into different segments.

Disks have 2 main properties
Seek time - measured in ms - this is a factor for RANDOM ACCESS RPM - Revolutions per minute - this is a factor for SEQUENTIAL ACCESS

The OS is a few seeks at bootup (Random Access to the start of each OS File), then a few random read/writes (seeks) for registry changes, paging file, windows logs, etc The Transaction logs are Sequential Writes (RPM-based) The Database is a LOT of Random SEEKS (Seek time better than RPM)

If you partition a disk to separate the OS from the transaction logs - registry writes, windows logs, etc happen, then a MAJOR disk seek to move the head to the other part of the disks.

If you partition a disk to separate the Transaction Logs from the Database - sequential write to the transactions logs, then a MAJOR disk seek to move the head to the other part of the disk.

Failures can happen on a partition (a logical arrangement of the disk) but more than likely a failure happens to the disk itself - so all partitions on that disk are lost during a failure.

With 500GB SATA Drives less than $60 these days - and Terabyte disks under $100/ea - it's worth the investment in more disks.

To give you a real performance example - in business-grade servers (like an HP DL380 - my standard building block for Exchange Servers) - I can put in 8x 2.5" SAS drives. Why do we use 2.5" SAS instead of 3.5" SAS drives - the platters are so much smaller - they significantly reduce the physical movement of the drive head - so they reduce seek time.

We are also keeping an eye on SSD (Solid State Drives) as they have nano-second (instead of milli-second) seek times - I just had a discussion with Scott Whigham about them recently.

I hope this helps.