Disk technology
Your disk capacity requirement is a very important input to the disk configuration planning process. This process contains many factors.
This is not a simple process that involves many other factors including:
- Survivability
- Configure the disks with the ability to survive single or multiple disk failures (e.g., RAID-1 or RAID-10).
- Configure the disk array with multiple I/O paths to the server to survive I/O path failures.
- Configure the disks to be accessible from multiple server nodes to tolerate a single node failure.
- Manageability
- If you have very short time windows to backup the database, select disk arrays that allow you to take logical backups (e.g., array snapshots).
- Scalability/performance
- Configure the disk array with many small disks instead of a few large disks so that you can increase the number of I/O paths.
- Configure the disk array with large NVRAM cache to improve read and write performance.
- Configure the disks with stripping (e.g., RAID-0 or RAID-10).
Let's take for example that you need 900GB and you have disk arrays or storage area networks (SAN) that are made up of 93GB disks. The following table summarizes the trade-off choices for the common disk organizations. Let's further assume that the database is implemented over ninety 10GB data files.
| Tech | Scalability | Survivability | Maintainability | Num Disks |
|---|---|---|---|---|
| JBOD | Poor - Subject to throughput of individual disks | Poor - Single disk failure creates outage and require database recovery | Poor - High disk utilization skew | 10 |
| RAID-0 | Excellent - Striping N disks provides read/write throughput at N times a single disk | Poor - Single disk failure creates outage and require database recovery | Excellent - expect near uniform disk utilization within a logical unit. Potential LUN utilization skew. | 10 |
| RAID-1 | Poor - similar performance to JBOD | Better - Could survive multiple disk failures in different mirrored sets | Poor - High disk utilization skew | 20 |
| RAID-5 | Excellent for read - similar to RAID-0. Potentially poor for write performance. | Better - Able to survive a single disk failure. Multiple disk failures creates an outage and require database recovery. | Excellent - low disk util skew. Possible LUN utilization skew. | 11 |
| RAID-6 | Excellent for read (similar to RAID-0). Potentially poor for write performance as parity calculations need to happen. The performance of RAID-6 and RAID-5 is about the same and dependent on the controllers. | Better. Can survive a double-disk failure. This gives it an edge over RAID-5. A failure of more than two disks creates an outage. | Excellent. Low disk utilization skew. Possible LUN utilization skew. | 12 |
| RAID-01 | Excellent read/write performance. | Could tolerate up to two disk failures as long as both failures are not in the same mirrored set. | Excellent - low disk util skew. Possible LUN utilization skew. | 20 |
| RAID-10 | Excellent read/write performance. | Could tolerate up to N disk failures as long as there isn't two failures in a mirrored set | Excellent - low disk util skew. Possible LUN utilization skew. | 20 |