Delivering as promised, ProLiant's screws were spring-loaded and accessing the inside of the box was truly a pleasure. It was only the fact that we had it loaded a single-faced, 19-inch rack that caused some problems: The side panel that lets you access the server's RAM was blocked by the rack's rails. We reseated the DIMMs to address our high ECC level, figuring that they might have loosened during shipping, and discovered that we couldn't remove the panel without pulling the unit completely forward. You'll definitely want to plan on installing this unit in a cabinet. If cache expansion is a priority, then you'll want to look at other vendors' products; the 5500R offers only a 512-KB CPU cache, not a 1-MB cache.

CIM, like NetFrame's Maestro, has a Windows-based client in addition to SNMP support. The client lets you connect to any Compaq server on your network, check LCD alerts and browse log events remotely. We were impressed with CIM's completeness. It tracks power supply events, potential memory failure, Automatic Server Recovery, environment problems, network card and SCSI array; Micron's and ALR's servers are similar, but they're missing Automatic Server Recovery. We'd like CIM to include a pager utility like that of ALR's Revolution.

Compaq's ProLiant definitely wins for beefiest power supplies; at 700 watts per supply, these power supplies will drive anything that you install on the server. We liked the separate power cords for each power supply, which the ALR server did not include. In addition to letting you plug into separate electrical circuits, the separate power supplies provided some necessary physical distance from the power path, contributing to this unit's fault tolerance.

Although some of Compaq's higher-end units--such as those in the 6500 and 7000 series--support hot-pluggable PCI, the ProLiant 5500R does not, which made it fall short of NetFrame's fault-tolerance standard.

ChatCom ChatterBox
The ChatterBox that we received was a combination of cabinet rack and server. The server as a whole is the rack, with consolidation hardware, server modules and impressive power supplies festooned all over the cabinet. The price of the entire system (comprised of two servers, one management station and tremendous storage) was almost three times the cost of the other systems. The server's performance didn't justify its $55,000 price tag.

The system, cabinet and other components shipped with two server boards, each housing a maximum of two processors. The servers from ALR, Compaq and Micron also came equipped with two processors, but with expandability to a minimum of four processors. ChatCom's product doesn't work in tandem with its two server boards. Although you could hot-swap the CPU boards from the backplane, popping out one of the dual CPU board brought down the entire server. None of the servers we tested exhibited hot-swap CPU capabilities.

We were disappointed at having to modify I/O cards so we could put them in the server. Modification consists of gripping an acceptable I/O card by the bottom quadrant of its bracket with a pliers and bending it to a 90-degree angle. This is cumbersome and doesn't speak well for the system's design. Even NetFrame's FICs, with their impressive fault-tolerance features, don't require user modification of I/O cards.

Although ChatCom's system shipped with an external cached RAID array--a repackaged Storage Computer Corp. RAID device--it still rated second to last in our network performance tests, indicating that there's definitely more to network performance than just I/O.

Also, ChatCom's server module finished last in the Neal Nelson & Associates' memory and CPU tests. ChatCom said its ability to connect many servers to the same RAID device was important to users; however, if your shop hasn't bought into the consolidated server idea, this may not be important to you.

The ChatView/IM management software was installed on a separate management PC module on the ChatterBox. It was attached to ChatCom's proprietary IM (Intelligent Management) bus and monitored the server. ChatView/IM has a Video Select status, Phone Status and a deadman reset and timer, which is perfect for a consolidated remote PC rack, but not as informative as we'd like for server monitoring. ChatCom should emulate Compaq's potential trouble-reporting capabilities, though its deadman reset was a welcome feature.

ChatterBox's most troubling aspect is its price of $55,000. Unfortunately, because it was designed for consolidated computing, ChatterBox offers features--such as the ability to plug in up to seven server modules in the same chassis--that weren't particularly relevant for our testing this time around. The external RAID device also contributed to ChatterBox's hefty price tag, but because the unit didn't perform extraordinarily well, we felt that the device's cost wasn't justified.

Jonathan Feldman is the technical systems manager for Chatham County, Savannah, Ga. He can be reached at jonathan@co.chatham.ga.us.

How We Tested Rack-Mounted Servers
To test each rack-mountable server's raw computing power, we used Neal Nelson & Associates' Business Benchmark, which measures relative performance of several systems and isolating specific system components. Each server was configured identically, and the tests ran while the servers were disconnected from the network.

To evaluate network performance, we used Network Computing's FileMetric 1.0 written by Coffee Computing Corp. (www.coffeecomputing.com). It's a multithreaded client/server master/worker software package that benchmarks file server performance using a group of client machines. The file server can be any combination of hardware/software that can serve files to Windows95 or NT clients. FileMetric tests file reads, writes and seeks across SMB (Server Message Block) file systems, each of which can produce multiple test engines.

We used 12 Fast Ethernet engines running on Pentium clients on an idle segment to hit each server with the same test data and methods. This testing included multiple reads and writes, as well as small and large application launches. We performed multiple runs for each test and averaged the results for a mean of read and write throughput. We configured each server with 256 MB of RAM, 512 KB of cache, two processors and Windows NT 4.0 with Service Pack 3 and ran identical network protocols and services.