Scaling Windows NT Server Storage and Fault Tolerance

ID: Q170921

The information in this article applies to:

Microsoft Windows NT Server versions 3.5, 3.51, 4.0

SUMMARY

This article discusses using Windows NT Server's software fault tolerance capabilities to scale invested hardware to meet current and future storage demands. Note that these capabilities are not as fast or dependable as hardware fault tolerance but can help extend existing hardware functionality.

This article is primarily for systems administrators who have installed systems that use hardware RAID arrays and who want to combine two or more hardware RAID arrays as one functional logical drive to expand space, improve performance, and extend the usefulness of existing hardware investments.

MORE INFORMATION

Scaling multiple RAID arrays can exceed the bandwidth of the BUS that the SCSI controllers are using, so knowledge of SCSI capacity planning is assumed in this article.

For more information regarding SCSI capacity planning, please see the Microsoft TechNet article "SCSI: Hardware Performance Considerations for Win95/NT" under Hardware/Hardware Products Architecture.

IMPORTANT! Introducing Windows NT's software fault tolerance requires that you keep a valid copy of the DISK key by either having current backups of the registry and a recent emergency repair disk, or using the Disk Administrator's Partition/Configuration/Save option to set up a valid DISK key backup.

SCALING METHODS:

NT Mirroring Hardware Stripe Sets (sometimes referred to as RAID 10)


Disk0---(Stripe)----DiskX
      ^
      |   (mirrored)
      v
Disk0---(Stripe)----DiskX

Pros: The fastest RAID array method; excellent read and write speed.

Cons: Mirroring provides the minimum FT required. Any corruption of the data may replicate to the shadow set. This provides no integrity protection for the drive arrays because they are running a stripe set underneath. Any drive failure would render the one of the arrays offline.

H/W RAID-5 Arrays in NT Volume Set


Disk0---Disk1---Disk2---DiskX (Hardware RAID 5 array)  member-1
      (NT Volume Set)
Disk0---Disk1---Disk2---DiskX (Hardware RAID 5 array)  member-2

Pros: Allows very large logical drives with RAID 5 fault tolerance underneath to protect the volume set. Allows two dissimilar RAID 5 arrays to be grouped together. This allows you to concentrate on the size of the logical driver rather than on performance.

Cons: Performance of the RAID arrays is not combined as RAID 10. This gives the least-optimal performance of all the configurations. Additionally, the logical or physical loss of one member results in the total loss of the entire volume set.

NT Mirroring of H/W RAID-5 Sets


Disk0---Disk1---Disk2---DiskX (Hardware RAID-5 array)
      ^
      |   (mirrored)
      v
Disk0---Disk1---Disk2---DiskX (Hardware RAID-5 array)

Pros: Very good read performance; good write performance. Both the data and drive arrays have fault tolerance.

Cons: Any corruption of the data could be replicated to the shadow drive. Very costly due to duplicate hardware requirements.

NT RAID-5 on Multiple H/W RAID-5 Arrays


Disk0--Disk1--DiskX (H/W RAID-5 array)\ 
Disk0--Disk1--DiskX (H/W RAID-5 array) |-RAID-5 created in Disk Admin.
Disk0--Disk1--DiskX (H/W RAID-5 array)/

Pros: Excellent read speeds; good write speed. This is the one of the best combinations of speed and security. The array and the data have the benefit of RAID protection.

Cons: The number of h/w RAID arrays required to make this efficient is a disadvantage. Three RAID-5 arrays in a soft RAID-5 would lose one-third total space to parity. Generally desirable would be four or five arrays to make the parity space required more reasonable (one-fourth or one-fifth total space).

Additional query words: prodnt scale fault tolerance

Keywords : kbsetup kbhowto nthowto NTSrv
Version : winnt:3.5,3.51,4.0
Platform : winnt
Issue type :