Radified Guide to SCSI - Boot from a SCSI Hard Drive

Conclusion Major Premise The Argument Argument-part2 Minor Premise
Seek/Access Tekram/Adaptec This vs that Misc Misc config
Misc config 2 Config/Compare Comparison Linkage Home

Seek / Access Times

Two things make the SCSI interface superior to IDE for running an operating system, applications, and swap file:

1. dramatically lower low seek/access times (3.6ms seek for SCSI vs 8.5ms for IDE - user benchmarks vary)
2. SCSI's multitasking, multi-threaded capability (IDE is a single-tasking interface)

In order to read a file, the disk actuator must first move the read/write heads to the correct cylinder. This is referred to as positioning, and is measured by a metric called seek. (typically 4 to 10 millisecs). Then the read/write heads have to wait for the correct sector, containing the desired data, to come spinning around. This time is solely dependant on spindle speed, and is referred to as latency. The faster the drive spins, the quicker the correct sector will arrive at the read/write heads, and the lower the latency. 

Sometimes the correct sector will be right at the read/write heads, and no waiting (latency) will be involved. Other times, the correct sector will have just passed the heads, and it will take a full revolution for the correct sector to arrive at the heads. Typically, latency is averaged .. equal to one half the time it takes to make a full revolution. Seek + latency = access.

Some people have the (mistaken) idea that IDE/ATA drives would be just as fast as SCSI drives if manufactures would only offer 10- or 15Krpm IDE/ATA drives. This is false. A drive's seek time has *nothing* to do with spindle speed (RPM). If an IDE/ATA drive spun at 1,000,000,000 rpm, it would still have a slow seek time, cuz seek is determined by how long it takes the read/writes heads to find the correct cylinder, once it receives the command. 

People who think that IDE/ATA drives would be as fast as SCSI drives if they were offered/manufactured with 10/15Krpm spindles show that they doesn't understand how a a hard drive works. The performance of the drives *actuators* are what determines how quickly the read/write heads find the correct cylinder. Spindle speed merely affects latency and sustained transfer rates, which aren't as important as seek (for running an operating system, applications & swap/page file).

I once read that, in the world of hard drive technology, one millisecond is an eternity. If that's the case, then current generation IDE drives are several eternities slower than their SCSI counterparts. SCSI drives are more than *twice* as fast as IDE drives at getting to the data once a command is received. 

The seek time spec is technically a function of the drive, and not influenced by the controller, or operating system per se. These things may have a perceived effect (benchmarks), but, technically, seek times are solely a function of the drive itself.

This is what gives a SCSI-based system that wonderful feeling of responsiveness that make SCSI users so fanatical. It's hard to describe if you've never experienced it. When I upgraded my system/boot drive from 7200rpm IDE/ATA (IBM), to a 10Krpm LVD SCSI (also IBM), I had great Wow! factor .. not unlike when I first got a Cable modem, or my first 3D accelerator. It was dramatic. From a seat-of-your-pants perspective, it felt as if my system ran 3X faster

By definition, (average) access time = (average) seek + (average) latency. Latency is determined (solely) by how fast a disc spins, and can be derived mathematically. Average latency for a 7200rpm HDD is 4.2ms, and 3.0ms for a 10Krpm drive, and 2.0ms for a 15Krpm drive. Let's compare access times (seek + latency) for the fastest ATA & SCSI drives.

As you can plainly see, the X15 accesses data over twice as fast as the fastest ATA100 drive out there. This does not even take into account SCSI's multitasking/multi-threaded capabilities (which the ATA interface does not sppt). Perhaps you can see now, why a SCSI drive is so much faster than the fastest ATA drives.

IBM has recently announced a 15Krpm drive - the 36Z15 - with manufacturer seek/access specs of 3.4/5.4ms.

You don't have to know how & why this is so, but it helps to understand how things work. I have copied the contents of an interesting 5-point post I found at an online bulletin board. Note that the following five points are not my own, and I forget who posted it, but it's obvious that the gentleman knows what he's talking about, and says it more eloquently than I could. You may find it interesting. Paste begins here:

  1. In general, real-world performance of a modern disk drive under a modern operating system using a modern file system is far more dependent on data access time (RPM and seek) than sustained transfer rate (density) - regardless of whether the system is a workstation or a server. You gain very little improvement in performance as density (STR) increases.

    In modern OS, such as Windows 2000, multiple threads do I/O independently and virtual memory system accesses paging file in stochastic manner. In addition, modern file systems such as NTFS does fragment, and uses transactional logs, permissions, etc. All these leads to disk access patterns that are random in nature.

  2. Because of #1 above, IDE drives do not scale well as load (number of I/O requests) increases. Meaning that - if you put a non-trivial load, such as multitasking, on an IDE drive, your system will slow to a crawl.

  3. Improvements in density (STR) far outstrip improvements in access times. In the last 3 years, density has increased by ~800% (double every year), while data access times improve by less than 100% (far less for IDE drives - seek time has improved very little for IDE drives over the past few years). The difference is now so large that a modern hard disk drive spends most of its time seeking, and not actually transferring data.

  4. Winbench is largely a lightly-loaded, single-tasking benchmark. It does not accurately represent a real-world environment, where people typically perform non-trivial tasks. As such, it biases heavily against SCSI drives. For example, go to storagereview.com, and compare the latest Maxtor DiamondMax Plus 40 against the 2-year old Seagate Barracuda 9LP under Winbench and under Intel IOMeter.

    The Maxtor beats the Barracuda in Winbench by a large percentage, but the 2-year old Barracuda beats the Maxtor decisively in Intel IOMeter workstation test.

  5. It's cheaper for manufacturers to improve density than access times, because higher density means less (number of) platters, and therefore, lower cost. Improvements in sustained transfer rates (STRs) is more or less a by-product of increased density. Thus, it's in the manufacturers best interest that consumers perceive STRs as the most important factor in determining performance - when in fact it is not.

    SCSI drives generally emphasize data access times, and that's also why Seagate's Cheetah X15 uses very small platters and can only store 18 Gigs on a low-profile drive while IBM's new 75GXP IDE drive, with a much higher density, can store 75 Gigs.


The gist of these points is that the access time spec is more important than sustain transfer rates, when it comes to running an operating system, and that SCSI drives have much better access times than their IDE counterparts. Note that no one is saying that IDE/ATA drives, or the IDE/ATA interface, sux. It doesn't.

IDE has its place in every PC (I'm currently running 3 IDE hard drives), but they do not perform (nearly) as well as SCSI drives at running an operating system.

The StorageReview says here that "...it's evident that random access dominates typical workstation usage ... Though the loading of executables, DLLs, and other libraries are at first a sequential process, subsequent accesses are random in nature. Though the files themselves might be relatively large, parts of them are constantly being sent to and retrieved from the swapfile. Swapfile accesses, terribly fragmented in nature, are quite random

Executables call other necessary files such as images, sounds, etc. These files, though they may represent large sequential accesses, consist of a very small percentage of access when compared to the constant swapping that occurs with most system files. Combined with the natural fragmentation that plagues the disks of all but the most dedicated defragmenters, these factors clearly indicate that erring on the side of randomness would be preferred."

And here, SR adds, "...STR had relatively little effect upon overall drive performance. Today, it should be clear that steadily-increasing transfer rates have in effect "written themselves out" of the performance equation ... it should be clear that random access time is vastly more important than sequential transfer rate when it comes to typical disk performance. Thus, the reordered "hierarchy" of important quantifiable specs would read:

The Storage Review is generally considered an authority in hard disk drive (HDD) technology, performance, & benchmarking measurements, and educators in the field of drive technology. There is much to learn about HD technology by reading their reviews & reference material, and hanging out at their general bbs forum - inhabited by some of the more brainier online types. They have the great IDE/ATA vs SCSI debated honed to a fine science.

Next -> [SCSI Guide - Tekram vs Adaptec]

Previous -> [SCSI Guide - Minor Premise]