Tuning I/O resources

Performance considerations for RAID 4 and 5

An extra performance consideration for RAID 4 and 5 configurations is that their I/O throughput becomes progressively worse as the ratio of the number of writes to reads increases. If the ratio of writes to reads is very high, the performance of RAID 4 and 5 can drop below that of a RAID 1 (mirrored) array.

RAID 4 and 5 arrays perform poorly on writes because of the extra disk operations required to update the parity information. The operating system performs the following sequence of events when writing a cluster to a RAID 4 and 5 array:

  1. Read the original cluster data.

  2. Read the original parity.

  3. Subtract the original data from the parity.

  4. Add the new data to the parity.

  5. Write the new parity.

  6. Write the new data.
As a result, a write requires a total of four disk accesses and two calculations on the parity.

Three-piece RAID 4 or 5 arrays use two pieces for data and the remaining piece for parity. This allows an optimization to be made which avoids reading the parity piece. The sequence of events performed by the operating system when writing becomes:

  1. Read the data piece other than the one to be updated.

  2. Generate parity from other data piece and new data.

  3. Write the new parity.

  4. Write the new data.
In this case, the write involves three disk operations and one parity calculation. To take advantage of this optimization, we recommend that you use three-piece RAID 5 arrays when constructing a RAID 53 array.

Next topic: Choosing a cluster size
Previous topic: Tuning virtual disk performance

© 2003 Caldera International, Inc. All rights reserved.
SCO OpenServer Release 5.0.7 -- 11 February 2003