DISK BENCHMARKING

5/1/98

Disk Benchmarking: There are two general approaches to disk benchmarking. One is to measure data on basic disk parameters. This is not especially easy. Moreover, it is hard to extrapolate the measured values to actual disk performance. The second approach is to set up a test that attempts to emulate typical disk usage. Both types of test are in use.

Basic disk parameters include rotation speed, data transfer rate, and head switching time.

Rotation speed is important because moves to an arbitrarily selected location on the disk will result in an average delay of half a rotation. This is called "Rotational Latency". Disk rotation rates vary from about 3600 rpm in the mid 1980s to perhaps 7200 rpm for high end 1998 drives. The value is usually determined from the drive specifications although it is possible for software to measure it.

Disk Transfer rate is the ability of the disk to transfer large blocks of continguous data. Provided that the proper controller is used, and the data transfer mode on the motherboard is properly set, DTR will limit the speed of the disk once a data transfer starts. DTR is measured in Megabytes per second and varies upwards from a fraction of a megabyte per second upwards to perhaps 50 megabytes per second. DTR is likely to vary between different areas of the disk, and computed rates may be affected by caching or lack of it, etc. Disk benchmarking programs attempt to compensate for these effects, but don't always succeed. Minimum and maximum values reported by benchmarks often apply only to small areas of the disk with special properties. For example, they are likely to be highly misleading with older drives if the drive elects to pause to recalibrate in the middle of the measurement.

Head positioning times vary with how far the head has to move. They are usually obtained from the specifications. Either average or track to track times of both may be specified. Average times may vary from about 85 mSec in 1985 to 10 mSec or less in 1998. Head positioning has minimal affect on long sequential transfers but is likely to be the most important factor in loading Windows software which often requires finding and loading a large number of small files.

Measurement of all the above parameters is complicated by the presence of caching on the drive and in the software as well as by CMOS settings.

The typical performance approach to Disk Benchmarking is typified by Ziff-Davis WinBench which uses a mix of disk operations from typical office software.

Popular Benchmarking programs include:

Benchmark Description
H2benchA Free DOS program with extensive capabilities.
Adaptec ThreadmarkFree for WIN95 or WIN NT
WinBenchFrom Ziff-Davis Pblications. Attempts to test "typical usage. Some results from WINBENCH97 may be suspect. A small fee is charged for the CDROMs.
WinTuneFrom Windows Magazine. Free and easy to use, but results are not thought to be especially accurate.
BurnCWIN95/NT measures DTR. Free for non-Commercial use
HDTachWIN95, Basic Version is free. Write testing requires purchase.
WIN95 Performance Monitor SYSMON.EXE.Can be used to compute overall DTRs. Free.

Return To Index Copyright 1994-2002 by Donald Kenney.