Hard Drive Basics 4

The Meaning of Benchmarks

Measuring Hard Disk Drive Performance

How do you go about choosing one particular drive over another? You would do best to base your decision on a wide range of criteria. First off, you will need to determine the manufacturer's reputation for product quality and reliability. Then, there are the drive's performance specifications. Manufacturers use a number of common performance specs. Since hard disk drives combine both mechanical and electronic functions, there are specs to measure hard drive performance in both of these areas.

Unfortunately, there are such a great number of specs that manufacturers have a tendency to cite only a few of the available specifications, giving only a partial picture of a given drive's performance. Most commonly cited specifications - seek time, head switch time, rotational latency, and access time - measure electromechanical performance (the performance of mechanical functions that are controlled electrically). Disk and host (IDE - SCSI) transfer rates reflect the performance of electronic functions. Another performance mark - data throughput - measures the rate of data transfer between the disk and host.


Seek Time
This is the amount of time it takes for the actuator arm to move the read/write head between tracks. The ability to move from the current track to the next requested bit of data might entail moving just one track or up to as many tracks as there are on each platter. Seek time is measured in milliseconds and between adjacent tracks can be as short as 2 ms, while a full-stroke seek - movement between the outer and inner track - can be as much as 30 ms.   Average seek time - the time it takes to position the drive's read/write heads for a random request - on most of today's drives ranges from 8 to 14 ms.

Seek time, while constantly quoted by drive manufacturers, defines but a single aspect of drive performance. Unfortunately, as convenient as this performance mark is, it is not necessarily an accurate representation of actual drive performance. Since it measures only a part of random drive operation, use seek time as only one of many benchmarks for evaluating any given drive.


Head Switch Time
On multi-platter drives, the actuator arm moves all of the read/write heads over the platters synchronously. Since only one of the read/write heads can be in use at a time. Measured in milliseconds, head switch time is the average amount of time the drive takes to switch between two of the read/write heads when reading or writing data.


Cylinder Switch Time
All of the tracks at the same position on different platters together make up a cylinder. A cylinder switch, also called a track switch, requires a movement of the actuator arm to position the read/write head over another data cylinder. Also measured in milliseconds, cylinder switch time gauges the average time the drive takes to switch from one cylinder to another when reading or writing data.


Rotational Latency
Once the head is positioned over the proper track, it must wait for the drive to spin the platter to the correct sector. This wait state, called rotational latency, is measured in milliseconds and depends on how fast the platter is spinning. The longest of these wait states occurs when the head reaches the track just after the sector has rotated past the head's location because it then must wait a full platter rotation before it is correctly positioned over the proper sector for data transfer. On average the disk needs to make 1/2 rotation before the next sector to be accessed is under the head. So, we can see that the higher the RPM the smaller the latency period.

One of the ways manufacturers are reducing rotational latency is to increase the disk's rotational speed. But remember, faster disk rotation generates greater heat and friction - which causes wear. Rotational latency can also be reduced by employing cylinder skewing. Cylinder skewing is a technique that attempts to position the next logical track to be accessed under the read/write head at the proper time. During the time the read/write head takes to move between tracks, the beginning of the track might have rotated past the read/write head. Having arrived too late, the read/write head stays in position until the drive finishes the rotation of the platter. It begins its transfer as soon as the beginning of the track rotates under the head. To avoid these long periods of rotational latency, the cylinders are skewed, that is, the beginning of each track is staggered to allow for the time the drive takes to move between them.

Still another method, head skewing. works on a similar principle. During a head switch, the disk platter continues its rotation. Once the drive completes the head switch, the read/write head attempt to start a read (or write) of the next logical sector after it has rotated past. Arriving too late, the read/write head must stay in position until the drive finishes the rotation of the platter. It begins data transfer as soon as the appropriate sector rotates under the head. By staggering the locations of the sectors across the platters, head skewing ensures that the next logical sector of data to be accessed is positioned under the read/write head immediately after the drive completes a head switch.


Data Access Time
Access time is time, in milliseconds, it takes to position a read/write head over a particular track and find the required sector within the track for data transfer. Data access time is a combination of seek time, head switch time, and rotational latency. Albeit still incomplete, data access time is a much better indication of a given drive's performance than just seek time.


Data Transfer Rate
Once the heads are properly positioned, the drive is ready to transfer (read or write) data. This entails a data transfer between the disk and the CPU. The faster the data transfer, the less time the user has to wait for a software program to perform the function for which it was created.

Data transfer depends on two operations: the disk transfer - how fast data is passed from the disk to the hard drive's controller - and the host transfer, or the speed at which the controller passes data to the CPU. The data transfer rate is measured both in megabytes (amount) and megabytes per second (speed). In order to increase the host transfer rate and minimize mechanical delays (seek time and rotational latency), manufacturers have added cache memory buffers to the hard drive's PCB. A cache buffer provides temporary storage that reduces bottlenecks in the flow of data between the disk and CPU. These bottlenecks occur when the CPU has to wait for data from the much slower electromechanical disk, or because the disk drive controller transfers information faster than the operating system can respond. A cache buffer, when implemented along with an effective caching algorithm can increase the data transfer between a drive and the CPU by a factor of from 50% to as much as 100%.


Data Throughput Rate
Throughput rate amalgamates both data access time and data transfer rate, representing the total amount of data that the CPU can access in a given unit of time. Because of this, it is a reasonably comprehensive measurement factoring together most of the major drive performance measurements. Data throughput scores also indicate the speed of the host computer, so, as such, should not be looked upon as the only indicator of drive performance. Often you will see Data Throughput Rates included in advertisements for hard drives but it is important to remember that any measurement that does not include host data transfer time is theoretical since data throughput cannot be tested without connecting the drive to a host computer. Provided other system components can process the data as fast as the hard drive can read or write it, data throughput can provide a good indicator of disk drive performance. Data throughput is measured in kilobytes per second.

1998/1999/2000
MediaTek All Rights Reserved