Mirrored Stripes or Striped Mirrors

This is an excerpt from the bestselling book Oracle Grid & Real Application Clusters.  To get immediate access to the code depot of working RAC scripts, buy it directly from the publisher and save more than 30%.

This RAID level is often misunderstood. Which should be used, RAID 10 or RAID 0+1? Examine Figure 5.10 and see exactly how they work. In the case of RAID 0+1, also called mirrored stripe, striping is implemented in the lowest position, meaning from a group of drives. A stripe is formed and then two such stripes are mirrored resulting in a logical drive. With this method, the loss of any drive makes the entire stripe invalid and the stripe is detached from the mirror pair. When the failed disk is replaced, the entire stripe must be brought up to date. Recovering the entire stripe can take a substantial amount of time as all the stripe members need to update the data.

Figure 5.10: Mirrored Stripe and Striped Mirror RAID architecture

With RAID 1+0 (or 10), mirroring is implemented at the lowest position. A group of mirrored drives are used to create a stripe. RAID 10, also called a striped mirror, is the superior method. If a disk fails in a striped mirror layout, only the failing disk is detached, and only that portion of the volume loses redundancy. When the disk is replaced, only a portion of the volume needs to be recovered. As shown in Figure 5.10, mirrored drives are independent units. When a disk is lost, it affects only one mirrored pair. Thus, compared to a mirrored-stripe, a striped-mirror offers more tolerance to disk failure. If a disk failure occurs, the recovery time is shorter for a striped-mirror layout.

A short summary of the RAID levels is presented in Table 5.1.

RAID LEVEL

STRENGTHS

WEAKNESS

RAID 0

Performance

No redundancy

RAID 1

Redundancy without Parity

Cost (double the disks)

RAID 3

Minimal Write penalty

No Overlapping

RAID 4

Overlapped small I/Os

Parity Disk bottleneck

RAID 5

Overlapped small I/Os

Write Penalty

RAID 0+1/10

Redundancy and Performance

Cost is very high

Table: 5.1: Strengths and Weaknesses of Various RAID Levels

One thing to remember when calculating the number of disks required for a particular RAID configuration is not just the storage capacity required, but to also take into account the needed I/O requirements. On some configurations the added parity writes and other overhead associated with RAID5 for example, can reduce the I/O capacity of the drive set by up to 50%. Disks are limited to around 110-120 I/Os per second maximum I/O capacity for linear read/write activity. For random read/write activity this can drop to 90 I/Os per second or less. Taking I/O capacity into account, RAID10 or 01 becomes less costly with the modern RAID controllers since it essentially doubles I/O capacity even if it halves storage capacity.

Software RAID and Hardware RAID

As previously stated, the RAID operations can be performed either in the host or within the disk sub-system. When RAID is done within a host, usually with the help of volume manager software or by means of a device drive, it is referred to as software RAID. When it is implemented within the storage system it is said to be hardware RAID.

Some of the issues surrounding the software RAID implementation are:

* Issue of Portability ? Since the RAID software implementation has some O/S specific components, these components have to be RAID operations, share the kernel mode components, and may add to the system CPU load. Software RAID uses more system resources, as more disk ports and channels are required, and it is subject to additional loads during write and copy operations.

* Software RAID is relatively complex. Creating several dozen redundant performance volumes across several dozen hard drives results in several hundred configuration records that describe the layout.

The following advantages may be seen with hardware RAID:

* The RAID firmware is executed on a dedicated processor within the disk subsystem, and therefore does not share the system?s CPU.

* It is portable across all the operating systems. In the event of a malfunction in the RAID firmware, the host system continues to operate and gives a suitable report on the RAID issue. At the same time, if the crash occurs at the system level the storage system functioning is unaffected.

* Many of the RAID solutions are equipped with battery backup modules that allow them to maintain cache coherency and complete outstanding operations without loss of data integrity.

* RAID controllers or storage processors are specialized for enhancing performance. Auxiliary processors are dedicated to calculating the parity of the data blocks that are being written to disk, while the main embedded processor concurrently fetches and executes RAID code.

However, there is one situation where software RAID becomes very useful. To mirror the drives from two different storage units, only software mirroring at the host level can do it. For example, after drives from two storage units are placed under the control of the Veritas volume manager at host level, a volume can be created by using the RAID-1 level to mirror two disks originating from two different disk systems. By this means, even if access to one of the storage units is lost, the volume can still continue to function.