|
 |
| ||||||||||||||||||||||||||||
 |
sections
home features from the editors opinion interview toolkit issue index news forums about us about queue advisory board media kit about ACM privacy policy writers guidelines feedback 
|
Inside a Drive TodayThe cylinder, which was such an important allocation unit when there were lots of heads per cylinder, is fast losing its utility, even if it could be exposed through the interface. One of the most interesting trends in drives is the rapid reduction in the average number of disks in each drive. Today most drives—dominated by the personal desktop market—ship only a single disk or platter. In fact, a good percentage of them (about 31 percent industry-wide, according to estimates by Seagate market research) ship with a single head. These drives demonstrate one benefit of the terrific increase in areal density. For the cost-sensitive PC market, a capacity sufficient to satisfy most customers can be achieved most economically by using only one recording head on one side of a single disk. The most disks ever put into a 1-inch-high 3.5-inch disk drive is six, with a corresponding 12 heads. The difference between what is possible in theory and what customers buy is dramatic, as shown in Figure 3. Figure 3
Other factors are further reducing the significance of the cylinder dimension. The long-term trend toward smaller-diameter disks reduces the length of average tracks and shrinks the size of a typical cylinder. In particular, the highest-performance SCSI drives have adopted smaller-diameter media. In a 3.5-inch form factor the largest media possible is 95 millimeters. Although it offers maximum capacity, larger media has several negative side effects. It wobbles more, has more curvature both radially and circumferentially, and consumes more power for a given RPM. These make an increase in areal density more difficult to achieve. For these and other reasons, drives use smaller media as the RPM increases:
More recent changes in data layout effectively dissolved the cylinder concept. Logical blocks used to be ordered on a drive so that all sectors on one cylinder were used before going to the next cylinder. This is no longer necessarily the case. Spiraling the sectors along a single surface in one recording zone before moving to another surface in that zone has some advantages, such as sequential transfer performance. Consider the outer recording zone in a three-disk drive. The first n tracks of the drive would be the n tracks that make up the first recording zone on the first surface. The next n tracks of the drive would be the n tracks on the next surface, but starting at the inner diameter working outward to minimize seek length and preserve sequential performance. The next n tracks would be the same as the first n, but on the third surface, with the next n again going out from the inner track of the recording zone to the outer radius. (See Figure 4.) Figure 4
There are variations on this serpentine format, and not all drives use it. The point is that our notion of the cylinder as a fixed location of the actuator that would be desirable to use as an allocation unit is not a dependable concept. The layout of tracks themselves has undergone a transformation. Historically, each sector of user data has been framed by additional information that enables the drive to locate the right one and ensure that it is correct. This extra data used to consume about 20 percent of the total bits on a track. Figure 5 illustrates the important fields. Figure 5
The address, known as a header, was used to compare with the desired address to locate the target data. The error correction code (ECC) fields provided error recovery information. The gaps were needed to give the head time to turn the writer off or on as needed. Without those gaps, noise from the head could wipe out data or a subsequent sector header field. Because these fields represented overhead and loss of usable capacity, they were candidates for change so that more capacity might be made available to the user. When drives used inductive heads, every sector had its own separate header, as previously described. Magneto-resistive (MR) and giant magneto-resistive (GMR) heads made the headers a more complex problem. Although they had an MR element for reading data, they still had an inductive writer. These two elements doing the reading and writing were next to each other on the head. Because they were on a rotary actuator, they could not both be positioned over a desired track at the same time. This meant that the read element was not on the track when the head was positioned for writing. Drives often used two headers: One was in line with the data sector to validate the address for reading; the other was offset from the data sector to enable the address to be validated when writing. This was clearly going in the wrong direction, consuming more capacity instead of less. As TPI increased, servo information was placed on the data tracks instead of on a separate dedicated servo surface. Typical drives today might have three to five sectors per servo burst, depending mostly on whether the drive is SCSI or ATA. (More servo bursts consume capacity but provide more performance and reliability in the presence of external influences such as vibration.) These servo bursts took the form of short blocks of information that consume from 5 to 10 percent of the bits on a track. As drive designers developed experience with the embedded servo bursts, they realized that no headers were really needed. They could use the positioning information in the servo and just count until they passed enough sectors to arrive at the desired one. (See Figure 6.) Figure 6
| ||||||||||||||||||||||||||||||||||
Previous Page (3/7)