Managing RAID on Linux - Derek Vadala [24]
Disk Access Protocols
The disk protocol of the hardware you choose has a tremendous impact on the performance and scalability of your array. Each protocol has its own hard limits on the maximum throughput of each I/O channel and the maximum number of devices you can attach to a single channel. So the disk protocol you select will have a direct impact on the maximum size of your array.
Although we traditionally think of RAID in terms of high-end SCSI systems, today it's not uncommon for consumer-marketed systems to come equipped with support for RAID on non-SCSI disks. In fact, Linux software RAID can support either SCSI or ATA devices as part of an array (see the following section). The kernel will even let you mix these protocols within a single RAID device, although that arrangement isn't recommended. (See the Matched drives section, later in this chapter, as well as the previous section, I/O Channels.) Software RAID under Linux does not rely on the underlying disk architecture to work, so there is no reason why an array could not be built using a Firewire (IEEE 1394), Fiber-channel, or other disk architecture developed in the future, as long as you can find hardware and device drivers to support the architecture as a standalone device.
ATA (used interchangeably with the acronym IDE) and SCSI are discussed in detail throughout this book because they are the most common disk protocols in use today. ATA is a part of every modern motherboard, and SCSI is the most common choice for large servers.
The AT Attachment (ATA) and Integrated Disk Electronics (IDE)
Integrated disk electronics, or IDE, has had many incarnations and many names since its introduction in 1986. Originally, hard drives were small enough, in both size and capacity, to fit directly onto disk controllers. As storage requirements grew, manufacturers realized that housing drives on controller cards was an inefficient use of space. Soon drives and controllers became separate entities, connected by ribbon cable. This meant that drives could grow in size without interfering with the expandability of the motherboard. It was common for these integrated controller cards to make adjacent slots on the motherboard inaccessible. Manufacturers eventually decided that portions of the controller could be housed directly on the drives and that creating a standard drive interface would allow for both expandability and portability. Originally called IDE in several proprietary implementations, a standardized version called the AT Attachment, or ATA, was eventually ratified (although many people still use the terms IDE and ATA interchangeably). This new disk interface was called the AT Attachment because it was introduced with the ISA (AT) motherboard. It quickly grew in popularity, and today the ATA interface is the most widely deployed consumer disk interface. Figure 2-20 shows the ATA interface.
Figure 2-20. The ATA interface separated the drive and the controller.
ATA has evolved a great deal since its introduction. Its performance and scalability have improved over time. Table 2-4 outlines the various iterations of ATA.
Table 2-4. Overview of IDE/ATA types
ATA type
Maximum throughput (MB/s)
Common names
ATA-1
8.3
ATA, IDE, Fast ATA
ATA-2
16.66
EIDE, Fast ATA-2
ATA-3
16.66
Ultra DMA, Ultra ATA
ATA-4
33.33
Ultra ATA/33
ATA-5
66.66
Ultra ATA/66
ATA-6
100
Ultra ATA/100
ATA-7
133.33
Ultra ATA/133
Many of the names associated with various iterations of ATA represent departures from the ATA specification by a single manufacturer. Enhanced IDE (EIDE), for example, was an attempt by Western Digital to increase its market share by offering enhancements to the original ATA (ATA-1) specification before