These days, it seems I see more and more problems related to hard disks. This isn’t primarily on servers, although we’ve had our fair share of volume corruption, but on workstations.
The Problem
Users call in and complain about uncorrelated slowness, unresponsive applications, and even the inability to log into the domain when their password and network connection are great. Typically the Application log will be full of regular and frequent application or GPO errors. In the worst cases the System log will have NTFS, disk, and ftdisk errors, which is an absolute sign of disk problems.
Diagnosis
Beyond CHKDSK, there’s not much else the way of volume and partition checks built into Windows itself. You can determine the hard drive manufacturer and download their disk tools as well to check the physical state of the disk. The problem can be either logical (address by CHKDSK or repartitioning and formatting) or physical (click, click, click). Hopefully, it’s never the latter.
Recovery
You can run CHKDSK until you’re blue in the face, and it may never solve your problems. Instead, back up you data first. Then you can use DISKPART to CLEAN the partition scheme and recreate it and format the drive. This will address MBR and MFT corruption, which is first a logical issue. You may even want to use vendor supplied tools to completely wipe the drive clean before re-creating partitions and volumes.
This is also a great time to do some research on disk partition alignment.1
But, if the problem is the underlying physical disk, you’ll need to get a new disk before putting data back. Sure it sounds simple, but this all comes down to a large time expenditure, because you’ll be schlepping bits from either install DVDs or a backup drive. And that can take a while.
—
1 This was a hot topic for SQL Server about a year ago, but it’s relevant for any situation. By default, Windows 2003 and earlier buffer disk partitions with 63Kb. This causes a mis-alignment at the logical level. Instead of one logical read equating to one physical read, a portion of the logical layout is in one physical sector, while the rest is in another, requiring two physical reads of the medium. See the link above for a more detailed explanation.
