Reviews Leaderboard Database Reference Search StorageReview Discussion Reliability Survey Search About Contents

Second Anniversary!
3/13/2000 - SR`s IOMeter Tests
3/14/2000 - The New Database

Operating Systems and Benchmarks - Part 3
  March 13, 2000 Author: Eugene Ra  

ZD WinBench 99 Explained

WinBench 99 actually delivers two distinctive sets of disk measures. The first features a set of three separate low-level tests that measure drive parameters such as sequential transfer rate, average random access time, and CPU utilization. The higher-level Disk WinMarks, as mentioned above, are disk access patterns of popular applications isolated for repeatable playback.

The Business Disk WinMark 99 consists of disk access routines of the Business Winstone 99 in isolation. The Business Winstone 99 consists of three popular office suites (Corel WordPerfect Suite 8, Lotus SmartSuite, and of course, Microsoft Office 97) along with Netscape's Navigator. According to, "following the lead of real users, [Business Winstone 99 keeps] multiple applications open within each suite, and switches tasks between those applications and Netscape Navigator." Thus, the Business Disk WinMark 99, again just the disk access pattern of the Business Winstone 99 in isolation, reports only one aggregate score. The reported score is the average speed attained by the drive subsystem taking into account all factors: the disk's speed, the controller's speed, the operating system's cache settings, and the machine's memory subsystem. Thus, the Business Disk WinMark 99 scales upwardly as a system's "externals" such as CPU and RAM increase in speed.

As one may expect, the High-End Disk WinMark 99 consists of the disk access routines of the High-End Winstone 99 in isolation. Its applications include AVS/Express 3.4, FrontPage 98, Microstation SE, Photoshop 4.0, Premiere 4.2, Sound Forge 4.0, and Visual C++. Unlike the Business Winstone, the High-End Winstone runs applications serially. This means that individual scores for each application are presented as well as a weighted average. The reported scores are the average speeds attained by the drive subsystem taking into account all factors: the disk's speed, the controller's speed, the operating system's cache settings, and the machine's memory subsystem. Thus, the High-End Disk WinMark 99 scales upwardly as a system's "externals" such as CPU and RAM increase in speed. It's important to note that here one may often witness KB/sec figures in excess of an interface's limit (say, 66 MB/sec or 80 MB/sec). Such high figures may be indicative of the applications extensive use of disk caching beyond the capabilities of the drive itself.

WinBench 99's Disk/Read Transfer Rate measures the sequential transfer rate across an entire partition, quantitatively reporting the measured result at the Beginning and End of the partition as well as outputting a GIF file that graphically represents STR across all cylinders. Previously we've simply reported the Beginning and End values, but recently we've been noticing significant differences in zone sizes where one disk may maintain its highest level STR along a greater percentage of its total capacity than another. As a result, we'll include the GIF in all future drive reviews.

The Disk Access Time measures random access times across the partition. Remember not to confuse the manufacturer's specified seek time with access time (as so many users and, unfortunately, hardware sites do). The WinBench Disk Access Time test will take into account seek time, rotational latency, and any other overheads in reporting an aggregate score.

Disk CPU Utilization reports the % of CPU cycles utilized in transferring data. There are two settings, one to measure utilization at the maximum STR of the disk, and another to measure at a constant transfer rate, the default being a relatively sedate 4 MB/sec. We've standardized on the latter to maintain consistency across drives. Note that while we've received consistent CPU utilizations scores on our old testbed with Windows 95 and Windows NT, the combination of our new testbed and Windows 2000 has yielded somewhat erratic results. We recommend not relying on this figure too highly; instead, if CPU utilization interests you, look to IOMeter's measures (see below).

Preparations to run WinBench 99 include partitioning the drive into a single extended partition of maximum size and the creation of a max-size logical drive, formatted with NTFS, within said partition. This results in drive "E:" ready to be tested. Note that WinBench 99's "Test Settings" option under the edit menu must be changed to test any drive other than that on which WinBench is installed. A setting here may also be toggled to save a GIF of the Disk/Read Transfer Rate test. Note the importance of creating a single, maximum-size partition. Too often we've noticed others creating a partition that doesn't encompass the whole disk and therefore reporting incorrect average access times and transfer rates.

The five disk tests outlined above consist of one "trial." Three trials are conducted for each drive, with repartitioning, reformatting, and a cold boot occurring between each.

WinBench 99 is our "old faithful," but is getting a bit gray behind the ears. We've noticed certain instances (say, the Maxtor DiamondMax Plus 40 vs the Seagate Cheetah 36LP) where scores reported don't seem to correlate to our personal experiences. Hence the need for another good, corroborative benchmark: Enter IOMeter.

 Introducing Intel's IOMeter


Copyright © 1998-2005, Inc. All rights reserved.
Write: Webmaster