We rated the drives for speed, features, and ease of use. Our custom performance suite consisted of five components: sequential throughput, random access, CPU utilization, text searching, and video playback.
Our low-level CD-ROM tests were a portion of NSTL's InterMark performance benchmarks. In each test, drives were instructed to read data from inner, middle, and outer tracks of the CD. We accessed data from these three areas because CDs store information sequentially on a single track that's spiralled across the CD. Drives spin the CD faster when reading data from an inner track compared to outer tracks, and a drive's performance can vary depending on the location of the data on the CD.
The throughput tests measure the amount of data a drive reads into memory in a second. Data was read sequentially in a series of 16-KB blocks
. In the random-access tests, drives read a single 16-KB block then skipped over a third of the CD to read a second 16-KB block. The CPU utilization tests measure the percentage of CPU resources used to sequentially read 16-KB blocks at a rate of 150 and 300 KBps.
The text-search tests consisted of a single, keyword search and a Boolean search using Gofer, an information exchange protocol developed by the University of Minnesota. The text test data was ``Project Gutenberg,'' electronic text of important English language documents published by Walnut Creek CDROM (Walnut Creek, CA). In the first text test, we browsed a 5-MB file for a keyword. The second test used a Boolean search phrase to locate two words.
In addition, we assessed each drive's ability to play video sequences using Microsoft's VidTest CD-ROM video tests. The drives ran an 8-bit, 912-frame file in resolutions of 320 by 240 pixels.
EASE OF USE
We considered the quality of documentation, how easy it is to install the dr
ive's hardware and software components, and rated each drive's front-panel indicators.
The features that we considered most important were the length of the warranty, range of device drivers, MPC Level 2 certification, and support for SCSI-2 and fast SCSI. We also looked for the modes supported, such as CD-Audio, CD-ROM XA (Extended Architecture), CD-R (CD Recordable), CD-I (CD Interactive), and ISO-9660.
We ran the tests on a Compaq Deskpro 66M (66-MHz 486DX2) with a 500-MB hard drive and MS-DOS 6.2 and Windows 3.1. We disabled Smartdrive, and no swapdrive was used under Windows. We connected each drive to the test system using an Adaptec AHA-2740/42 SCSI (EISA) adapter and Adaptec EZ-SCSI 2.04. Three drives--Hitachi's CDR-1900SPC, MacProducts USA's Magic CD Pro, and Reveal's The Entertainer--use a proprietary SCSI connection; we used that for testing. Test files resided on the test computer's hard drive. Parallel-port drives ran from an AST PowerExec 4/25SL notebook with an enhanced parallel p
ort, 8 MB of RAM, and a 200-MB hard drive.
Contributors
Alan Joch, Senior Editor/BYTE, coordinates the combined testing between the BYTE Lab and NSTL.
Chandrika Mysore, Project Manager/NSTL, has tested peripherals and systems for NSTL since 1989.
Illustration: WHAT THE TESTS MEAN
RANDOM ACCESS
A measurement of how quickly a drive can read data stored across a CD-ROM (lower test scores indicate faster performance). This is important for interactive applications, including multimedia encyclopedias and electronic training programs, where data isn't easily prefetch
ed in the buffer.
CPU UTILIZATION
An essential gauge for video applications. It measures how long the system CPU is used to transfer data from the drive buffer to system memory (lower test scores indicate better performance). Too much time devoted to data transfers results in dropped frames.
SEQUENTIAL THROUGHPUT
The single most important performance test for CD-ROM drives because data on CDs is stored sequentially (higher test scores indicate faster performance). Sequential throughput is important when copying large files, and loading software from a CD.
TEXT SEARCHING
An application-level test that measures how many seconds a drive takes to find one and two keywords in a large text file (lower numbers indicate faster performance). Along with sequential throughput and random-access scores, this test indicates real-world performance.
How We Tested
