Configuring RAID for Optimal Performance

Overview Configuring RAID for Optimal Performance

1. Overview

The target audience for this guide includes users, technical support personnel, and pre-sales

personnel who work with Intel

Server RAID products. It is assumed that the reader has a basic

understanding of hard-drive operation, RAID operation, and RAID levels.

This guide provides information to help achieve optimal performance of a RAID array depending

on the RAID level and performance measurement tool used. It does not include techniques for

fine tuning RAID performance for a specific application.

This guide covers Intel

RAID Controllers and Intel

Integrated RAID only. This guide does not

cover Intel

Embedded Server RAID Technology II.

2. Performance Measurement Tools

When measuring RAID performance it is important to understand how the measurement tool

works, the capabilities of the tool, and its limitations. It is better to avoid using a tool with

unknown or unclear data access patterns, which can make the results difficult to interpret.

2.1 Copying large files

Copying files with the Microsoft Windows* and Linux* standard OS utilities generates primarily

sequential stream of data with 64 KB blocks and no queuing. Read Ahead (Always or Adaptive)

and Write Back modes must be used to achieve good performance with this type of operations.

2.2 Tools without queuing

There are many tools available for measuring storage performance, but many of them do not

have queuing capabilities and they can measure sequential throughput only.

Without queuing, Read Ahead and Write Back caching modes must be used to achieve good

throughput. Tools without queuing should not be used for measuring the maximum throughput

capabilities of a RAID controller because the RAID cache causes a throughput bottleneck. This

is especially important for read operations.

Some tools allow the access block size to be changed. If a large block size can be selected, this

can partially mitigate the lack of queuing. For example, sequential throughput with 2 MB block

size and no queuing may be similar to performance with 64 KB block size and a queue depth of

32 IOs.