不错的RAID内容

Cluster Hardware Considerations for Oracle RAC

This section covers some of the issues that you should consider when choosing hardware for your Oracle RAC system. Two of the key issues are how to choose the right RAID level for your shared storage, and how to choose the right amount of memory for your RAC servers.

RAID Systems

Managing a large number of individual disk drives can be very difficult, since you must balance the database files across all of these disk drives in order to spread out the I/O

load. In order to simplify this task, provide for optimal performance and to provide a fault

tolerant system, RAID systems were developed. RAID stands for Redundant Array of Inexpensive Disks.

RAID systems are very configurable, depending on what your needs are. These different configurations have different performance and fault tolerant properties and are known as RAID levels. These RAID levels work differently but essentially serve the same purpose, to create a logical disk drive out of two or more physical disks. A logical disk drive or logical volume looks to the operating system and relational database management system (RDBMS) like a disk drive, but in reality might be the combination of many disk drives. RAID volumes are combinations of multiple disk drives configured in a RAID array to provide the desired performance and fault tolerant properties.

RAID 0

RAID 0 is considered a RAID level even though there are no redundant properties associated with this RAID level. A RAID 0 takes a number of disk drives and stripes them into a larger logical volume. By using RAID 0 you can combine or stripe multiple disk drives into what appears to the operating system as a single large disk drive.

RAID 0 works by taking the data in the logical volume and striping that data across the array. The data in the logical volume is broken down into what are known as chunks or stripes (depending on the vendor). These chunks are typically 64K, 32K or configurable in size. The chunks are then allocated to the physical disk drives in a round-robin fashion as shown here.

RAID 0 Advantages

RAID 0 Disadvantages

No overhead from RAID processing. So maximum performance is reached.

No fault tolerance. If a single disk drive were to fail all data would be lost.

All disk space is used.

In an RDBMS environment, it is not recommended to use RAID 0. In the event of a disk failure (and disk failures are probably the most likely type of failure to occur) all of the data or programs would be lost and you must recover from backup.

RAID 1 and RAID 10

RAID 1 is known as mirroring. With RAID 1 the entire contents of your disk drive has an exact copy on another disk drive, known as the mirror. With RAID 1 a disk drive failure is transparent to the user. If a disk drive were to fail, the mirrored disk drive immediately takes over. The term fault tolerance refers to the fact that the system can tolerate a fault, such as the loss of a disk drive and continue processing seamlessly.

RAID 1 Advantages

RAID 1 Disadvantages

Excellent fault tolerance. RAID 1 can tolerate the loss of a disk drive.

RAID overhead. When writing to the RAID 1 volume two physical I/Os are required, one to each disk.

Read performance is increased since reads occur on both disk drives.

RAID 1 is expensive since you must double the number of disk drives that you purchase.

RAID 10 or RAID 0+1 is a combination of RAID 0 and RAID 1. With a RAID 10 configuration disk drives are mirrored and then striped. Thus you can take advantage of the RAID 0 disk volume where you can increase space and performance as well as taking advantage of the mirroring properties of RAID 1.

RAID 10 Advantages

RAID 10 Disadvantages

Excellent fault tolerance. RAID 10 can tolerate the loss of a disk drive, or even the loss of all mirrors

RAID overhead. When writing to the RAID 10 volume two physical I/Os are required, one to each disk.

Read performance is increased since reads occur on all disk drives.

RAID 10 is expensive since you must double the number of disk drives that you purchase.

RAID 5

RAID 5 uses parity for fault tolerance. The advantage of using parity is that instead of having to double the number of disk drives in the system, you only have to add one disk drive to store the parity. RAID 5 uses parity, but distributes the parity among all of the disk drives in the RAID volume.

RAID 5 is very popular because it provides a fault tolerant solution at a relatively low cost. For the cost of one additional disk drive fault tolerance is achieved, but this is at a relatively high performance cost. In order to maintain the parity, when a logical write (a write to the logical volume) occurs a number of steps are required:

1. The parity and data disks must be read.

2. The new data is compared to the data already on the disk drive and changes are

noted.

3. A new parity is calculated based on step 2. 4. Both the party and data disks are written to.

So, for a single logical write, four physical I/Os must take place. When calculating the number of disk drives that are needed in your system, you must take into account the additional overhead due to RAID 5.

RAID 5 Advantages

RAID 5 Disadvantages

Fault tolerance. RAID 5 can tolerate the loss of one disk drive in the RAID volume.

RAID overhead. When writing to the RAID 5 volume four physical I/Os are required.

Read performance is increased since reads occur on both disk drives.

RAID 5 fault tolerance can only tolerate the loss of one disk drive in the RAID volume.

Striping provides for greater performance since there are multiple disk drives in the RAID volume.

In the event of a failure, performance is severely affected since all remaining drives must be read for each I/O in order to recalculate the missing disk drives data.

RAID Comparison

Each RAID level has its own attributes and performance characteristics as described above. The following table shows a comparison of those attributes and characteristics. As you can see, there are vast differences among the most popular RAID levels.

© 2002 Performance Tuning Corporation 8 © 2002 Performance Tuning Corporation 9

RAID

Level

Read

Performance

Write

Performance

Fault Tolerance

Cost

RAID 0

Good

Good

None

Low

RAID 1 and RAID 0+1

Good

OK

1 logical write = 2 physical I/Os

Excellent. Can potentially tolerate the loss of multiple disks

Highest

Requires that you purchase 2x disk drives

RAID 5

Good

Poor

1 logical write = 4 physical I/Os (2 then 2)

OK

Can survive the loss of 1 disk at severely degraded performance level

Best for fault tolerance

Recommendations for Oracle Component Placement

Oracle is sensitive to read performance and sensitive to write performance on the redo log files and on the archive log files. Thus, the following recommendations are given.

OS Volume The OS should be installed on a RAID 1 disk volume. It is

important that you do not need to restore/rebuild the OS in the event of a disk failure. This can be very time consuming and expensive. The OS will certainly fit on one disk drive, and RAID

5 is generally not supported in a 2 drive configuration. In addition, the Oracle binary files can be placed on this volume.

Redo Log Files The Redo Log files should be placed on a RAID 1 or RAID 10

volume. The I/Os to the Redo Log files are 100% sequential and 100% writes, thus RAID 5 is inappropriate.

Data Files The Data files should be RAID 10 if the I/Os are 90% reads or

less. If the I/O pattern is 90% or greater reads, then RAID 5 is OK. Again, your budget may help determine this.

Archive Log Files The Archive Log files can either be RAID 10 or RAID 5,

depending on your budget. Archiving might take longer if it is RAID 5.

By using RAID fault tolerant volumes, much pain and expense can be avoided in the event of a disk failure.