14.2 Choosing a Hard Disk

The good news about choosing a hard disk is that it's easy to choose a good one. Major drive makers such as Maxtor and Seagate produce high-quality drives at similar price points for a given type and size drive. When you buy a hard disk in today's competitive market, you get what you pay for. That said, we will admit that we avoid IBM and Western Digital hard drives because we have experienced severe reliability problems with both makes.

Manufacturers often have two or more lines of drives that vary in several respects, all of which affect performance and price. Within a given grade of drive, however, drives from different manufacturers are usually closely comparable in features, performance, and price, if not necessarily in reliability. Neither is compatibility an issue, as it occasionally was in the early days of ATA. Any recent ATA hard disk coexists peacefully with any other recent ATA/ATAPI device, regardless of manufacturer. The same is generally true of SCSI drives. All of that said, we use Seagate and Maxtor IDE drives and Seagate SCSI drives when we have a choice.

Use the following guidelines when you choose a hard disk:

Choose the correct interface and standards

The most important consideration in choosing a hard disk is whether to use ATA or SCSI, based on the issues we described in the preceding chapter. Once you make that decision, choose a drive that supports the proper standards. For more about ATA versus SCSI, see the upcoming sidebar.

ATA

Any drive you buy should support at least ATA/ATAPI-5 UDMA Mode 4 (ATA-66). ATA-100 drives sell for little or no premium, although some economy drives support only ATA-66, as much for market differentiation as anything. The truth is that the fastest current ATA drives cannot saturate even an ATA-66 interface, so ATA-100 is of no real benefit. ATA-66 and ATA-100 drives are backward compatible with earlier modes, including PIO modes, so either ATA-100 or ATA-66 is fine. Choose whichever is available in the size and price range you want.

SCSI

If you're buying new, or if performance is the top consideration, buy Ultra160 hard disks. They usually cost the same as drives that use earlier SCSI standards such as 80 MB/s Ultra2 Wide (U2W) and 40 MB/s Ultra Wide (UW). Ultra160 has at least the potential to provide higher performance, depending on the SCSI host adapter you use and how many drives share the channel. If you find a great deal on discontinued or used U2W or UW drives, they're worth considering for systems that have only one or two drives. As with ATA, the interface can transfer data faster than the drives themselves, so depending on your configuration you may not sacrifice performance by using older SCSI modes. If only one or two hard drives share a channel, U2W is probably fine, and if you have only one drive, even UW shouldn't hamper transfer rates. Purchase only SCAM-compliant drives.

ATA Versus SCSI The relative performance of ATA versus SCSI hard drives is hotly debated. Some argue that ATA and SCSI drives often use the same mechanisms, and the additional overhead of SCSI therefore means that ATA drives are faster. That's true as far as it goes, but it ignores some important issues: Load If you compare a 7,200 RPM ATA drive to an identical model with a SCSI interface under light loads, the ATA drive will probably benchmark as slightly faster, although not enough to be noticeable in a real working situation. But much of ATA's speed advantage is due to the simplicity of the ATA interface, and that simplicity incurs a penalty as load increases. Under moderate to heavy loads, particularly in multitasking environments, SCSI simply outperforms ATA. There's no question about that. That's why everyone uses SCSI drives on servers and workstations. Multiple hard drives If your system has two or more hard drives, SCSI has a big advantage. ATA does not permit simultaneous I/O on a channel, which means only one drive can read or write at a time. With SCSI, you can have many hard drives on a channel, and all can read or write simultaneously at full bandwidth if the channel is fast enough. Operating system If you're running Windows 9X, the advantage of SCSI over ATA is minimized. Many benchmarks that are used to show that SCSI is no faster than ATA are run under Windows 9X. Under Windows NT/2000/XP, the throughput and concurrency advantages of SCSI become apparent. Faster mechanisms The fastest ATA drives use the same head/disk assemblies as the slowest SCSI drives. If you need the highest possible performance, your only option is SCSI because the fastest HDAs are available only with SCSI interfaces. We use ATA drives if cost is a major issue, if the system is likely to be CPU-bound rather than disk-bound, and if the system runs Windows 9X. If none of those three is true, we use SCSI. If one or two are true, we decide based on other issues, such as using SCSI if we need to install many peripherals and ATA if we don't.

Buy the right size drive

It's tempting to buy the largest drive available, but that's not always the best decision. Very large drives often cost much more per gigabyte than mid-size drives, and the largest drives may have slower mechanisms than mid-size drives. So, in general, decide what performance level you need and are willing to pay for, and then buy a drive that meets those performance requirements, choosing the model based on its cost per gigabyte. All of that said, it may make sense to buy the largest drive available despite its high cost per gigabyte and slower performance, simply to conserve drive bays and ATA channels.

Choose the best rotation rate for your application

Rotation rate specifies how fast the drive spins. For years, all hard drives rotated at 3,600 RPM. Several years ago, drives that rotated at 5,400 or 7,200 RPM started to become available, initially for servers. This higher rotation speed has two benefits. First, a drive that rotates faster moves more data under the heads in a given amount of time, providing faster throughput. Second, the higher the rotation speed, the lower the latency. Nowadays, inexpensive ATA drives rotate at 5,400 RPM and mainstream ATA drives at 7,200 RPM. Entry-level SCSI drives rotate at 7,200 RPM, mainstream models rotate at 10,000 RPM, and high-performance models at 15,000 RPM. All other things being equal, higher rotation speed provides faster data access and transfer rates, but with correspondingly higher noise and heat. We recommend using 7,200 RPM ATA drives and 7,200 or 10,000 RPM SCSI drives for mainstream applications. Choose a 5,400 RPM ATA model only when cost is an overriding concern, and even then you'll save only a few dollars by buying a 5,400 RPM drive rather than a 7,200 RPM unit. Choose a 15,000 RPM SCSI drive only if getting the highest possible performance outweighs the 50% to 75% price differential versus a 10,000 RPM drive.

Give seek/access times heavy weight if you work mostly with many small files

Seek time is a measure of how quickly the head actuator can reposition the heads to a different track. Statistically, for a random access, the drive heads on average have to move across one third of the disk surface. The time they require to do so is called the average seek time. Once the head arrives at the proper track, it must wait until the proper sector of that track arrives under the head before it can read or write data, which is called latency. Average latency is one half the time that the disk requires to perform a full revolution. A 7,200 RPM drive, for example, turns at 120 revolutions per second and requires 8.33 milliseconds (ms) for each full revolution. The average latency is one half of that, or 4.17 ms. The sum of average seek time and average latency is called average access time, and is the best measure of a drive's access performance. Do not compare average seek time of one drive to average access time of another. Because average latency is a fixed value that is determined solely by the drive's rotation speed, you can easily convert back and forth between average seek time and average access time to make sure you're comparing apples to apples. For an entry-level ATA drive, look for an average access time of 16 milliseconds (ms) or less. For a mainstream ATA drive, 14 ms. For an entry-level 7,200 RPM SCSI drive, 10 ms. For a mainstream 10,000 RPM SCSI drive, 8 ms. For a high-performance 15,000 RPM SCSI drive, 6 ms.

Give data transfer rate heavy weight if you work mostly with large files

In most applications, data transfer rate (DTR) is less important to overall performance than average access time. DTR does become crucial if you work primarily with relatively few large files (sequential access) rather than many smaller files (random access). DTR is determined by several factors, the most important of which are disk rotation speed, cache size, and the on-board circuitry. When comparing advertised data transfer rates, be aware that there are several possible ways to list them, including internal versus external and burst versus sustained. The various transfer rates of drives are normally well documented on the detailed specification sheets available on their web sites, and less well documented in typical marketing materials.

Overall, the most important basis for comparison is the sustained transfer rate. Note that on drives that use more sectors on the larger outer tracks, transfer rates can vary significantly between inner and outer tracks. For example, a 10,000 RPM Seagate Cheetah SCSI drive has transfer rates of 49.1 MB/s on inner tracks and 63.2 MB/s on outer tracks. The average of those numbers, called the average formatted transfer rate, is a good yardstick. For an entry-level ATA drive, look for average formatted transfer rate of 14 MB/s or higher. For a mainstream ATA drive, 30 MB/s or higher. For a 7200 RPM SCSI drive, 35 MB/s or higher. For a 10,000 RPM SCSI drive, 50 MB/s or higher. For a 15,000 RPM SCSI drive, 55 MB/s or higher. Note that none of these sustained transfer rates is fast enough to saturate ATA-66, let alone U2W or faster SCSI. Even the peak transfer rates are within the capacity of ATA-100, U2W SCSI, and Ultra160 SCSI.

Rotation rate, average access time, and data transfer rate are all favored by drives with smaller form factors, and in particular those with smaller platters and higher data densities. This is true because it is easier and less expensive to run small platters at high speed than large platters, and because the smaller physical size of the platters means that heads need not move as far to access data on any portion of the platter.

Get a model with large cache if it doesn't cost much more

Disk drives contain cache memory, which in theory provides benefits similar to those provided by L2 cache on a CPU. Low-end drives may have 64 KB or 128 KB cache, whereas mainstream drives typically have 256 KB to 512 KB cache, and high-performance drives may have 2 MB or more. Some manufacturers sell the same model drive with differing amounts of cache, often indicated by a different letter on the end of the model number. In our experience, larger caches have a relatively small impact on overall drive performance, and are not worth paying much for. For example, given two otherwise identical drive models, one with 128 KB cache and one with 512 KB cache, we might pay $5 or $10 more for the 512 KB model, but not more. Adding cache is cheap, but it doesn't provide the benefits of a fast head mechanism and a fast rotation rate, both of which are more expensive to implement.

Make sure the drive fits your computer

All drives use standard width/height dimensions and screw hole positions to allow them to fit standard mounting locations. Drives for standard PCs are available in two nominal widths, named for the size of the platters they use. Each width is available in different heights. Together, the width and height describe the form factor of the drive, as follows:

5.25"

Some drives, typically of large capacity, use the 5.25" form factor. These drives actually measure 6" wide and come in three heights. Full-height devices measure 3.25" vertically, and are relatively uncommon nowadays. About the only 5.25" full-height drives you may encounter are very large capacity SCSI hard disks intended for use in servers. Half-height drives measure 1.625" vertically, and are far more common. A few 5.25" drives have been made in third-height form, which measure 1" vertically. Any of these drives fits in standard 5.25" drive bays. All cases except some low-profile cases have at least one full-height 5.25" drive bay, which can also be used instead to hold two half-height 5.25" drives.

Relative to 3.5" hard drives, 5.25" drives typically have slower rotational speed, longer seek times, and higher latency, all of which translate to slower data transfer rates. These performance drawbacks are true regardless of the capacity or interface of the drive. The one advantage of 5.25" drives is that their larger physical size allows packing in more and larger platters, which in turn means that 5.25" drives, particularly full-height models, can have much larger capacities than 3.5" drives. Although many 5.25" SCSI drives indeed have very high capacities, this is not the case with 5.25" IDE drives. Such drives, notably the Quantum Bigfoot series, are low-end drives that are commonly found in consumer-grade PCs. These drives gain no advantage from their larger form factor. One of the best upgrades you can make to a system is to replace one of these 3,600 or 4,000 RPM 5.25" IDE hard drives with a modern 3.5" 7,200 RPM drive.

3.5"

Most hard drives use the 3.5" form factor. These drives actually measure 4" wide and come in two heights. Most drives are third-height, or 1" high. Some high-capacity 3.5" hard drives use the 1.625" high half-height form factor.

Pay attention to how much current the drive draws

Here's one that few people think about, but that can be critical. A drive that requires only a few watts at idle or during read/write operations can easily require 30 watts or more when it spins up. Spinning up three or four ATA drives (or even one high-performance SCSI drive) may draw more current than your power supply can comfortably provide. Nearly all modern drives and BIOSes automatically support staged spin-up, whereby the Primary Master ATA drive (or Drive 0 on the SCSI chain) spins up first, with other devices spinning up only after enough time has passed to allow each earlier device to complete spin-up. However, not all drives and not all systems stage spin-up, so note the startup current requirements of a drive before you add it to a heavily loaded system. The current requirements of a drive are normally detailed in the technical specification sheets available on its web site.

Here are some things that you can safely ignore when shopping for a drive:

MTBF

Mean Time Between Failures (MTBF) is a technical measure of the expected reliability of a device. All modern ATA drives have extremely large MTBF ratings, often 50 years or more. That doesn't mean that the drive you buy will last 50 years. It does mean that any drive you buy will probably run for years (although some drives fail the day they are installed). The truth is that most hard drives nowadays are replaced not because they fail, but because they are no longer large enough. Ignore MTBF when you're shopping for a drive.

MTTR

Mean Time to Repair (MTTR) is another measure that has little application in the real world. MTTR specifies the average time required to repair a drive. Since nobody except companies that salvage data from dead drives actually repairs drives nowadays, you can ignore MTTR.

Shock rating

Drives are rated in gravities (G) for the level of shock they can withstand in both operating and non-operating modes. For drives used in desktop systems, at least, you can ignore shock rating. All modern drives are remarkably resistant to damage if dropped, but all of them break if you drop them hard enough.