Setting It Up
I didn't have to look far to see how the Lab could benefit from PowerSMP's consolidation capabilities. In a corner of the Lab sits a minitower system that functions as the Lab's PDC. Next to it is another minitower that the Lab uses as the BDC. The Lab also has a third system for various applications and media storage.
Tight spaces. Next to these computers is a small desk loaded with equipment, including the monitor, keyboard, mouse, printer, external tape backup, keyboard/video/mouse (KVM) switch, two external modems, and a new external ISDN modem. Mounted to the wall is a switched hub for connections to the rest of the building and to the Internet. Behind all this equipment are a half dozen power strips, a power supply, AC power adapters, a dozen serial and parallel cables, and network cables all jumbled together in a rat's nest. The Lab's NT servers and related equipment take up about 14 square feet of space.
With the PowerSMP and a four-shelf rack, I could have consolidated the Lab's network servers to roughly 4 square feet of space, conserved power, and simplified the network topology. I could have placed the switched hub on the top shelf; the monitor, keyboard, and mouse on the second shelf; the PowerSMP on the third shelf; and the printer and power supply on the bottom shelf. With the available PCI and ISA slots, I could have switched the modems to internal connections on the backplane. But because I didn't have a four-shelf rack, I could only imagine how nice an uncluttered workspace would be as I began setting up the PowerSMP.
Plug it in. Installing Windows NT Server on the first system was uneventful. Cubix includes mouse and video drivers and SCSI support for both the DP 6200 and the BC Triton processor boards.
However, I found that the second system's CD-ROM drive is not multiplexed (shared) by the backplane, although the floppy drive is. I overcame this problem by connecting the EIDE ribbon from the CD-ROM drive to each of the processor boards as I installed NT Server. Cubix is aware of this shortcoming and plans to correct the problem in the next product release.
Another problem I encountered was accessing and replacing the hard disks. The disks attach to brackets and lay vertically in a tray instead of horizontally as in a midtower. The space is very cramped, and I cut my hand trying to get my fingers into the tight space to disconnect the power connections and SCSI ribbon.
My suggestion to Cubix is to include connections similar to common RAID configurations, in which the disks simply slide in and lock in place. Another suggestion is to include a tape backup device in place of the two drive bays above the floppy drive.
How It Performed
For my performance tests, I used the Lab's standard configuration: a set of client machines on a 100Mbps Ethernet network divided into four segments that simulates the workload of multiple users. (For details about the Lab's test environment, see "The Lab's Test Environment.") I used Bluecurve's Dynameasure 1.5 as the workload engine. (For information about this product, see Lab Reports, "Dynameasure Enterprise 1.5," September 1997.) The combination of Dynameasure and the Lab's test environment simulates typical user workloads and provides quantitative benchmarks that you can use to compare hardware and software performance.
For a comparison test system, I used a brand-name server running NT Server 4.0 and Service Pack 3 (SP3), with quad 200MHz Pentium Pros, 512MB of RAM, four SCSI hard disks, and an Adaptec 10/100 adapter. I selected Dynameasure's Copy All Bi-directional tests because of the random order of 16 different transactions that copy compressed data, uncompressed data, binary files, text files, and image files between the server and the clients.
I tested a range of 5 to 100 users. The Lab test server performed well and produced a peak throughput of 3.6MBps at about 40 users. I conducted the same test concurrently across the PowerSMP 4000 subsystem processors and measured a combined throughput of 4.2MBps at about 36 users. Average response time for the test server rocketed to 50 seconds; the PowerSMP maintained an average of less than 2 seconds. The only performance drawback was that the test server could maintain support for all 100 users, but the PowerSMP could support a maximum of only 40 users.
Report Card
Overall, I liked the PowerSMP's advantage of consolidating network servers and centralizing management. In addition to reducing space used and improving topology, the PowerSMP provides a considerable cost savings for anyone upgrading a network. Computers with 200MHz processors and 130MB of RAM cost more than $2000, even if you're thrifty. You would need to purchase five systems, cords, cables, switches, and network adapters just to match the computing power of the PowerSMP. Of course, you could consolidate applications on one large server and spend up to $100,000 on a quad-processor system.
The PowerSMP was noteworthy. It was relatively easy to install and use across the network. The PowerSMP is not limited to a Microsoft environment; Cubix is targeting Oracle, Novell, and Lotus. Software drivers are available for NT, Windows 95, and Novell NetWare. The technical documentation is not for the casual reader, but it is useful for reference. New users will prefer a more user-friendly installation and configuration guide, with plenty of screen shots. The base package includes a 1-year warranty, and the company's Web site provides excellent product information and white papers.
End of Article
Prev. page
1
[2]
next page -->
