Test Specs

We rated these CD-ROM servers based on their performance, usability, features, technology, and price (all on a scale of one to five stars, except for price). We derived the performance rating by averaging the results from three equally weighted tests. Each performance test measured the systems' performance under various conditions that occur in a real-world network environment, including network and data contention.

The overall-rating score comes from a 65:20:10:5 weighted rating of performance, usability, features, and technology, respectively. Because performance is the overriding concern with servers, we gave i t the highest weighting in our overall rating. We judged system performance based on raw thro ughput.

Test Methodology

We connected each CD-ROM server individually to a 10-Mbps Ethernet network using the network interface adapters provided with each product. Each server was configured according to the manufacturer's directions. Setup included installing any provided software needed to make the server and the CD-ROM drives visible to the test network. When CD-ROM management programs were provided, we installed and evaluated them to gauge their ease of installation. However, the servers we tested provide network access without these tools.

Our test network comprised three Dell Optiplex Pro GXs that had dual 200-MHz Pentium Pro processors and 64 MB of RAM. The OS for all six systems was Windows NT Server 4.0. We attached each of the client systems to the network through an Ethernet hub and cables.

Our performance test software is a proprietary application developed by NSTL. Executed from a client system, it makes file-read requests from a drive mapped to a CD-ROM drive in the server system. The application reads a specified number of 16-KB sequential blocks of data from a contiguous 650-MB file on a specially authored CD.

Each read test is executed for a fixed period of time. Then the cumulative number of bytes read is reported and the performance, in bytes per second, is calculated. Thus, the tests measure throughput speed for sequential reads of one file.

No evaluation of random-read speed was conducted, as random reads are far less frequent than sequential reads. Because all files are written to the original CD-ROM in sequential fashion, there's no file fragmentation. In addition, each CD-ROM drive in the system under test was loaded with an identical CD-ROM disc to eliminate variances in performance due to file size or location on the disc.

Cache Evaluation

Our first performance test evaluates data transfer speed when reading uncached data. The CD-ROM server is rebooted to ensure that no data is cached. A single c lient system requests 1000 16-KB blocks of data from a single mapped CD-ROM drive. Because this is the first time the server reads the information from the CD, neither server- nor client-based caching has an effect on performance. Theoretically, the throughput for this test should closely match the maximum data-delivery speed of the CD-ROM drive itself.

The second performance test measures the efficacy of the CD-ROM server's caching. A second, different workstation reads the identical 1000 blocks of data from the same file on the same CD. We found that it was vital to use a different workstation here; if we used the same workstation, NT's caching completed the test with no network activity at all.

In this test, the impact of the server cache is isolated; the server should cache the data from the previous test. If the CD-ROM server has enough cache to hold the entire section of the file read in the test, the data transfer speed should approach the speed of the network interface. These two single-cl ient tests provide an indication of the CD-ROM's data transfer speed and the effects of server-side caching.

The third test maximizes contention for SCSI and network bandwidth so well that the TAC Systems HotSwap LanRedi TowerDrive was unable to complete it. We used a single-client system for this test and mapped each of seven CD-ROM drives on the server being evaluated to an individual drive letter. Then we started seven iterations of the test program simultaneously as separate tasks in separate sessions.

For this test, each program requested 2000 16-KB sequential blocks of data. To properly service these requests, the CD-ROM server must manage the data traffic across the network interface as well as the use of its SCSI channel. The total throughput for the test is the sum of the throughputs for each individual session.

On the fastest servers with the fastest drives, the aggregate throughput closely approaches the limit of network speed. The larger the number of CD-ROM drives, and the faste r each drive was (the systems came with 8X and 12X drives), the more likely the response time under high loads will be restricted by the pipeline.

In high-traffic environments, the network bandwidth of the CD-ROM server might be a factor when you choose your system. The four systems with specialized SCSI/network-interface-card (NIC) connectors -- the Boffin 7 Bay Tower, MDI CD-Express Connect, Microtest DiscPort Tower-7, and TAC HotSwap LanRedi TowerDrive -- offered only a 10-Mbps network interface. The two systems we tested that included full-blown servers, the Excel CDS-14 and Microtest Enterprise Group DiscPort Enterprise Server, supported 100-Mbps network adapters.

Several of the devices offer software that implements a strategy known as load balancing . Load-balancing software allows you to load several identical copies of a CD-ROM onto the same system. In a situation where many users require simultaneous access to the same files, the server automatically rolls over the requests to th e next free CD-ROM drive to avoid contention and reduce the overall waiting period for users.

---