REVIEWS

Low-Cost Servers: Easy To Use, Easy To Manage And Easy On Your Pocketbook

by Kiran Movva

Low-cost servers generally are used by departmental workgroups of 10 to 50 users and in those companies where the cost of the equipment is one of the major factors considered when making a purchasing decision.

Low-cost servers normally do not have the features you would expect to find in a higher-end solution, such as multiprocessor capabilities, flexibility in expansion and performance. Additionally, when your needs become more demanding, you will probably replace rather than upgrade these servers. Scalability is also important but isn't often found in lower-priced systems. Service and support become a more important piece of the puzzle in smal l office environments where a trained technician may not be present.

In this roundup, we evaluated six servers from Compaq, Dell, Hertz, Hewlett-Packard, IBM and Polywell. Each vendor was asked to deliver systems under $10,000, which at minimum included a 90-MHz Pentium or higher processor, 32 MB of RAM, two 2-GB SCSI drives, an Ethernet card and a CD-ROM.

What We Found The servers from established vendors such as Compaq, Dell, HP and IBM, were easier to set up, have bundled installation and management tools (except the Dell), and have less expensive service and support. Hertz and Polywell, on the other hand, offer custom-built servers shipped with options based on your needs. The customization also includes s erver management tools for an additional cost, but no server setup or network operating system installation aids. We felt the software installation and server management products should be bundled with the server. These tools really do save time, effort and reduce support costs in the long run.

Although Hertz and Polywell both use standard hardware components such as Adaptec SCSI controllers, we experienced failures on power-up (Hertz), PCI card detection problems in NT (Hertz), and operating system crashes due to hardware configuration (Polywell).

High marks go to Compaq for having the best value for the price. While it was not the most scalable of the servers tested, the ProSignia is still a complete server with above average performance, manageability, setup tools, and comprehensive service and support. HP and IBM also delivered solid products, but steer clear of the IBM if cost is an issue.

Compaq Computer Corp. Compaq ProSignia 300

The ProSignia 300 is a feature-rich package with above average performance, excellent manageability and low maintenance. With server setup and management tools superior to those provided by H P and IBM, the ProSignia represents the best value for the price.

Configuration software includes server setup and operating system installation aids. Compaq's management software represents the best of class with its Compaq Insight Manager. CIM allows for the remote management of Compaq servers, either via the network or via an asynchronous connection. In addition, SNMP and Pager support provides the features that IS managers are looking for. IBM and HP provide similar remote management solutions, but CIM provides the most detailed information.

Attention to detail is a mark of quality, and Compaq builds a solid low-cost system. Small features such as precise labeling of all ports and a cover that protects the power switch are probably not features that would make you go out and buy a Compaq. But these features do show Compaq's attention to detail. Besides the software tools that make managing and installing the server a breeze, the Compaq ProSignia is very well built and designed to me et future needs. One pitfall is that the ProSignia has only two PCI slots, while the IBM and Hertz provide four PCI slots for maximum growth potential.

Compaq's service and support standard options also are best of class. A three-year warranty, three years of next-day parts and labor, and enhanced phone support for the life of the server makes the Compaq a well-supported server. Only HP and IBM had similar structure for service and support.

Hewlett-Packard Co. HP NetServer 5/133 LC

While the HP has a solid, low-cost server with the NetServer 5/133 LC, it performed poorly, placing last among the servers tested. In addition, it is not as scalable as the other products, and the lack of free 24x7 technical support only detracts from this otherwise good server. These faults are uncharacteristic of HP's support reputation.

The HP ships with an Adaptec-2940 disk controller, as do most of the servers tested, and uses an Intel Ethernet NIC. We were surprised by HP's poor performance scores and could not identify the source of these problems.

The NetServer continues HP tradition in building quality products, providing such things as a swing-out cover for easy access. But this product lacks internal bays, and, unlike the Hertz, Polywell and IBM, it only provides for two PCI slots, limiting its scalability.

The HP comes standard with server setup and software installation aids, and includes server management tools. These help distinguish the HP NetServer from other vendors such as Polywell and Hertz and provide a complete package that eases installation and lowers support costs. HP was the only one of the established vendors not offering free 24x7 su pport. Instead, business-day support is provided for free, and 24x7 is offered as a priced option.

IBM Corp. IBM PC Server 320

The IBM PC Server 320 is the most expensive of the ser vers we tested. It demonstrates excellent scalability for future growth with plenty of free slots, drive bays and multiprocessor capability. However, for the price of the configured system, and the additional features delivered relative to other servers, we feel the Compaq ProSignia is a better value.

The PC Server had the best aggregate score among all servers. The high price and better performance is probably due to the 512 KB of write-back cache (other servers had 256 KB), Ultra Wide PCI and a larger case for additional drive bays and so on. Only the Polywell system had a larger footprint.

As with the HP NetServer, the IBM has lockable power and reset switches, with flash ROM-upgradable components like the servers from Compaq and HP. Flash ROM allows you to add support for new peripherals, fix code bugs and so on, simply by booting the server with the updated code. However, the 320 does not allow a hidden EISA configuration partition as the Compaq and HP servers do. This is important because it makes the tasks of adding new EISA cards and troubleshooting the system easier, since you do not have to keep track of EISA configuration disks or diagnostic utilities. There is nothing worse than trying to configure a system in the early hours of the morning only to be unable to locate your EISA configuration disk. Similar to the Compaq and HP servers, the PC Server 320 includes setup and management software, which rounds out a complete package.

Hertz Computer Corp. Hertz P166e

With good performance and scalability, the Hertz P166e will suit you as your workgroup grows. But if manageability, ease of use and setup are what you are looking for, the Hertz may not be for you.

The Hertz server had the second best performance rating, and is quite scalable for future needs. Similar to s ervers from Dell and Polywell, it lacks server setup and network operating system installation aids, making it more work for the system administrator. Additionally, it does not bundle any management tools nor support special partitions for diagnostic utilities.

It took us a great deal of time to get the server up and running. Windows NT was not able to auto detect any of the PCI cards that shipped with the server. After a few conversations with Hertz and unboxing a second machine, we were quickly on our way. The Hertz system was the only server with EDO RAM. Most of the servers shipped with ECC or Parity RAM. ECC RAM is considered to be more reliable, allowing for correction of memory errors, whereas EDO RAM is considered to be faster.

Like the Polywell server, management software may be ordered with the Hertz server for an additional cost. Hertz also charges additionally for next-day parts and labor support. Considering these additional hidden costs, this system prices out higher.

Polywell Computers Poly P6-200D1

Similar to the Dell PowerEdge and the Hertz P166e, the Polywell server lacks bundled setup and management tools, and service and support c ome at an additional cost. With multiprocessor compatibility-being able to hold 1 GB of ECC RAM-and with more than an adequate number of PCI and drive bay slots, the system is as scalable as the IBM PC Server 320 with enough room to meet the needs of most applications.

The Polywell took a while to set up-Windows NT crashed with IRQ errors. After receiving a second system, a Pentium Pro model, we were back on track. Polywell was the only Pentium Pro model tested. Although this did affect performance relative to other machines-all of which had a P5 chip-we did not see a spectacular spread.

While the Polywell server excels in scalability, it is not in the same class as the Compaq, HP and IBM servers, which offer such things as the various management tools and attention to detail. These are the features that make other servers easier to use and more reliable. If performance and scalability are of ultimate importance to you, and manageability, ease of use and setup are not, then the Polywell may be for you.

Dell Computer Corp. PowerEdge EL 5133

The PowerEdge EL 5133 includes good service and support, but doesn't include much else. The server does not ship any of the additional value found in our top performers, such as management software (management software is only available on Dell's high end XE and SE servers). And the Dell product is not as scalable as the other servers tested.

The Dell was the least scalable server, shipping with Parity RAM and only ISA bus slots free in the system. EISA or PCI bus slots are necessary for future expansion needs in a server-especially when you are adding peripherals. This leaves only ISA slots for additional cards: The idea of manually configuring ISA slots on a server is a scary thought. We would have liked to have seen either more PCI slots or an EISA bus, which offers better performance than an ISA bus. n

Kiran Movva is a systems analyst at a major energy corporation on the West Coast. He can be reached at kmovva@nwc.com.

Low-Cost Servers: How We Tested

While putting these servers through their paces, we wanted to simulate the daily activity that would be typical of office users. This was accomplished via a suite of application tests designed to simulate a bank of workstations connecting over a network to the server. To get a better sense of server performance, we also performed this same activity on the server, taking the network out of the equation.

For the first wave of testing, we simulated 10 workstations hitting on each server. The workstations involved in the testing represented a range of capabilities. We simulated two 486/66s, six P5/60s and two P5/90s. Additionally, each of these virtual workstations ran a copy of Windows from the server.

The suite involved the installation of nine applications on each server. Servers were tested with Microsoft NT Server 3.51 as the NOS. The applications installed were: 1-2-3, Freelance, Harvard Graphics (DOS), Wo rdPerfect (DOS), dBase(DOS), Paradox (DOS), Word, Excel and ccMail. For each of these applications, we implemented scripts that were comprised of common activities. For example, in Word, the recorded activity included a spell check, graphics paste, print preview and text formatting. In Freelance, tasks included slide show, printing and page down to name a few.

The first test was performed by attaching each workstation to the server, running Windows and randomly executing all nine applications and their activity scripts. Reported values are an average of the time it took each workstation to complete the test suite. Windows NT's Performance Monitor was used to measure CPU utilization throught application testing.

For the second suite, we installed Word, Excel and Powerpoint on the server and ran similar activity scripts. This time, we hoped to capture local server performance without network overhead-eliminating other network-related potential bottlenecks. Reported values are an average o f two cycles through our test suite. Again, Windows NT's Performance Monitor was used to measure CPU utilization.

Updated May 31, 1996