Best Overall: CD-ROM Servers

It's hard to make blanket statements about CD-ROM server performance. In heavy-traffic environments, CD-ROM servers that can single-handedly fill the Ethernet pipeline spend much of their time waiting for opportunities to transmit their data. But when the same small amount of data must be read repeatedly by various clients, cached performance is a more important consideration than total throughput.

The clear leader in our testing, both in the PC-based server category as well as overall, was the Microtest DiscPort Enterprise Server. Not only did the DiscPort Enterprise turn in the most impressive results in all three of our data-delivery tests, it also ranked highest in t he number of features offered. The device's lack of adequate documentation, however, lowered its usability score to somewhat below average.

Although it couldn't match the performance of the DiscPort Enterprise, the Boffin 7 Bay Tower nonetheless turned in top marks in the non-PC server category. Its data throughput never exceeded about 60 percent of that of the DiscPort Enterprise, but it was impressively consistent. Given its easy setup and bargain price, the Boffin 7 deserves a serious look for all but the most performance-critical applications.

Serving Data

In the uncached data test, we measured how fast each CD-ROM server could satisfy a unique request from a single client for a sequential file read from a single file on a single CD-ROM. In theory, the maximum rate of data delivery from the server to the client is constrained to either the CD-ROM drive's transfer rate or the bandwidth of the network, whichever is smaller.

All the sys tems that we tested used 8X or 12X SCSI CD-ROM drives, so the only external hardware-imposed performance limit was the Ethernet connection. Internally, your choice of SCSI and network hardware might create a performance bottleneck. The two PC-based servers, the DiscPort Enterprise and the Excel CDS-14, both came equipped with 10-/100-Mbps Ethernet NICs. The four non-PC-based CD-ROM servers we tested were each equipped with 10-Mbps Ethernet connections.

The DiscPort Enterprise delivered uncached data at 51 percent of the network's maximum capacity, which is a remarkable achievement indeed. The device bested its nearest competitor in this test, the Boffin 7 Bay, by 34 percent, and the MDI CD-Express Connect by a staggering 750 percent.

The cached-data test provides an insight into both the adequacy of the server's cache and the design of the CD-ROM-drive-to-Ethernet data path. A system with effective caching should be able to deliver the requested data at memory speed, subject to the bandwidth limit ation of the Ethernet port.

With the DiscPort Enterprise, DiscPort Tower-7, and Excel CDS-14, the cached-data performance did indeed jump to approximately 80 percent of Ethernet capacity. Meanwhile, the Boffin 7 Bay showed only a minor improvement over its uncached performance, which is symptomatic of a too-small cache. The CD-Express Connect exhibited no significant performance change compared to its uncached performance.

Juggling Conflicts

We designed the multiple-client/multiple-disc test to stress the server systems and evaluate their ability to manage contention for the network, internal SCSI bus, and cache. Again, the DiscPort Enterprise led the pack, delivering the same aggregate amount of data as it did in the single-client test, but divided essentially equally among seven clients.

The Excel CDS-14 and DiscPort Tower-7, both of which performed well in the cached-data test, fared poorly under the strain of serving multiple high-demand clients. These systems, which h ad previously managed to fill over 80 percent of the available network bandwidth when delivering data to a single client, dropped to 24 percent and 12 percent of network bandwidth, respectively, when attempting to serve multiple users. Again, the data that was delivered in our tests, though of reduced quantity, was approximately equally divided among the clients.

The TAC Systems HotSwap LanRedi TowerDrive, which scored well in the uncached single-client tests, would not run the multiclient test at all. This unit is based on the same Axis StorPoint technology as the Boffin 7 Bay Tower, but while the 7 Bay uses Plextor drives, the TowerDrive uses drives from Toshiba. Because TAC was unable to resolve the problems we encountered with this system, we gave the TowerDrive poor performance and overall-score ratings.

In the multiclient test, as in the others, the 7 Bay clung tenaciously to the middle of the performance range, filling from 38 percent to 49 percent of the Ethernet pipeline regardless of cac he status or number of users. Still, given the 7 Bay's aggressively low price, this moderate but consistent performance is entirely reasonable and a good value.