Managing RAID on Linux - Derek Vadala [29]
In general, it's a good idea to use the same disks in an array whenever possible. But using identical disks might not be an option all the time. Disks are made of several parts that affect their overall performance. If a situation arises in which you are forced to mix different disks, then you will want to know how to best evaluate a new disk to ensure that it will function appropriately when added to an array.
Platters, tracks, sectors, and cylinders
Two mechanical parts that affect performance are found on every hard drive. Inside each drive are magnetic platters, or disks, that store information. The platters, of which most common hard disks have several, sit on top of each other, with a minimum of space between each platter. They are bound by a spindle that turns them in unison. The surface of each platter (they are double-sided) has circular etchings called tracks, similar to a phonograph record, with the important difference being that tracks on a hard disk are concentric circles, while a record has a single spiral track.
Each track is made up of sectors that can store data (see Figure 2-24). The number of sectors on each track increases as you get closer to the edge of a platter. Sectors are generally 512 bytes in size, with some minor deviation that depends on the manufacturer.
Figure 2-24. The surface of a hard disk platter.
The speed at which these disks spin affects how fast information can be read. The rotation rate, or spin rate, of a hard disk is measured in revolutions per minute (RPM). Modern drives operate at speeds equal to or greater than 5400 RPM, with 7200 RPM being the most common consumer rotation rate. Drives operating at 7200 RPM are able to rotate through an entire track of data 120 times per second (7200 revolutions per minute / 60 seconds = 120 rotations per second). Most ATA drives spin at one of these rates, depending on the price of the drive. Older drives have slower spin rates. Faster SCSI drives, like those found in high-end servers or workstations, typically spin at rates of 10,000 RPM or higher.
Actuator arm
The second analog bottleneck is the actuator arm. The actuator arm sits on top of, or below, a platter and extends and retracts over its surface. At the end of the arm is a read-write head that sits a microscopic distance from the surface of the platter. The actuator arm extends across the radius of a platter so that different tracks can be accessed (see Figure 2-25). As the disk spins, the read-write head can access information on the current track without moving. When the end of the track is reached, it might seem logical for the actuator arm to move to the next track and continue writing. However, this would greatly increase the time needed to read or write data because the actuator arm moves much more slowly than the disk spins. Instead, data is written to the same track on the platter sitting directly above or below the current platter. A group of tracks, on different platters, that are the same distance from the spindle are called cylinders. Since the actuator arm moves every read-write head in unison, the read-write is already positioned to continue I/O. During a write, if there is no free space left on the current cylinder, the actuator arm moves the read-write heads to another track, and I/O resumes.
Figure 2-25. Actuator arms move heads across the surface of the disk.
When data is read or written along a single cylinder, and then along adjacent tracks, I/O is extremely fast. This is called sequential access because data is read from the drives in a linear fashion. When data is spread