Dual Pentium IIs pack power into servers

Testing by Andreas Uiterwijk and Chip Pettirossi

If you are looking for a PC server with plenty of horsepower and storage space, a 450 MHz Intel Corp. Pentium II dual-processor system may be the answer.

This class of workgroup servers is designed for file storage and data safety. However, the dual processors offer good investment protection because they let you add storage later or run applications directly from the server, which frees up desktop space. And these servers are affordable, priced at about $10,000 or less.

The newest PC servers on the market are powered by Intel's Xeon processors. However, most servers with Xeon processors are being offered only in four-processor configurations, which places them in the enterprise server category. For a workgroup file server roundup, the dual 450 MHz Pentium II processor configuration is more appropriate, which is why the FCW Test Center chose to test the Pentium IIs instead of the Xeons.

We tested offerings from four vendors: Compaq Computer Corp., DTK Computer Inc., Gateway Inc. and Micron Electronics Inc. We also requested servers from Dell Computer Corp., Hewlett-Packard Co. and Intergraph Corp., but these companies declined to participate in this review. (Dell told us its product was at the end of its life cycle, HP said the company was between product cycles, and Intergraph said its product was in transition.)

All the servers in this comparison feature Intel's 440BX chipset with a 100 MHz frontside bus, 256M of SDRAM and up to 32M of Redundant Array of Independent Disks cache. We asked for a RAID 5 disk configuration, which offers faster performance than other RAID configurations and protects data by striping it across all five hard disks. Therefore, the 40G to 45G of total storage space we specified was reduced to 30G to 35G of usable storage space.

To measure server performance, we ran Bluecurve Inc.'s Dynameasure/File Professional Edition 2.0, a benchmark that simulates users working on networked clients and servers. We reported the results using two measurements of hard disk performance: throughput and average response time. Throughput represents the input/output rate for file workloads, measured in kilobytes per second. Average response time is the average amount of time, in seconds, that each client request to the server was acknowledged (see benchmarks sidebar).

In all four servers, the performance bottleneck was the disk subsystem. In other words, I/O to the disks was where performance suffered when the machines were stressed. In all cases, CPU utilization was low because of the insufficient disk capacity to adequately "feed" the processors. Improvements to the disk subsystems, such as running more than one SCSI channel on the RAID arrays, probably would increase performance results for every server.

The test center also observed that server management capabilities have increased. Features such as intrusion detection and system utilization monitoring can head off trouble, and the latest management software enables you to check factors ranging from fan speeds to memory status to system and CPU temperature.

Compaq's ProLiant 1600 earned our Best Buy Award, with a final score of 8.15 on our scale of 1 to 10. It turned in the best average response time and the second-best throughput, and we loved the completely tool-free, modular chassis. Micron's NetFrame 3100 followed with a score of 7.14. It edged ahead of the Gateway ALR 8200 NTS because of its higher throughput and slightly better system management. The Gateway came in a close third with a score of 7.00 and the benefit of excellent expandability. DTK's APRI-81S/P450 was hurt by poor management, documentation and technical support, giving the product a score of 6.12. However, the DTK unit was by far the least expensive system in the roundup. At $5,480, it was $4,000 less than the least expensive of the other three.

* * *

Compaq's ProLiant 1600

Compaq's ProLiant 1600 turned in the best average response time in the review at 0.14 seconds and the second-best throughput score at 2,316 kilobytes/sec, which is so close to the Micron server's score that the difference is negligible. It was the only system to earn excellent scores for chassis design and system management. These factors combined to earn the ProLiant 1600 a first-place finish with a final score of 8.15.

The ProLiant 1600's expandability is good, with six expansion slots and nine drive bays. The expansion slots consist of two PCI slots and four shared slots, which give you a lot of freedom. Five of the drive bays are RAID-configured, and there are three standard 5.25-inch drive bays and one two-thirds-height 5.25-inch drive bay. The system features integrated graphics with 1M of graphics memory, an integrated 10/100 Ethernet controller and integrated dual-channel Ultra/Wide SCSI controllers with one internal port and one external port open. A cable already is connected to the internal SCSI port for any peripherals you may want to add. The RAID controller card features 6M of cache, which is the maximum for this controller.

Compaq blew us away with its innovative chassis design. Its completely tool-free chassis is modular and has a removable card cage assembly and CPU cage (including memory and fan). Say goodbye to those annoying, tiny screws because even the PC Cards feature tool-free removal. You can remove the top and side case covers separately, and the inside of the cover features a nice drive installation and configuration guide, complete with options installation and other configuration instructions. The unit features a front cover lock, but the power button is behind a spring-loaded door that is not lockable. The CD-ROM and floppy drives are located behind a clear plastic door, and when the cover is locked, they are accessible but cannot be removed. Compaq offers a tower-to-rack conversion kit that includes rails and a front bezel for $455.

Compaq's server management software, Insight Manager, allows you to monitor thermal characteristics, voltage, fan rpm, power supply status, memory status, processor activity, drive activity and physical intrusions into the case. Setting up Insight Manager is a bit more cumbersome than other management software, such as Intel's LANDesk Server Manager, but once installed, Insight Manager is fairly easy to use. Compaq also provides small utilities included in the server's control panel. These utilities, such as programs for checking card settings and status, were a great help in configuring the server for network use.

Compaq's RAID configuration management tool, Smart Array Manager, is available through the setup partition or on bootable floppy disks. It also includes a Microsoft Corp. Windows NT version. We found this software easy to use and intuitive. Checking the status of the RAID configuration is a breeze, and you can add

drives to the existing partitions on the fly. Overall, Compaq's RAID management tool was one of the better packages we saw in this comparison, and combined with the company's high-quality server management software, it helped earn Compaq an excellent management score.

Priced at $9,511 on the General Services Administration schedule, the ProLiant 1600 is the least expensive of the three brand-name servers in this review. Only the $5,480 DTK server is less expensive. Its tool-free, modular chassis design, excellent management and high performance scores place it a notch above the rest. This is a great overall package that includes very good support and a competitive price.

* * *

Micron's NetFrame 3100

Micron's NetFrame 3100 turned in a respectable second-place score of 7.14. It barely nudged Compaq's ProLiant 1600 out of first place with its top throughput score of 2,395 kilobytes/sec. However, its average response time was more than twice as slow as the Compaq server - and the slowest score of all four servers - at 0.35 seconds. Still, it finished slightly ahead of Gateway's ALR 8200 NTS because of its higher throughput and slightly better system management.

The NetFrame 3100 earned a good score for expandability, with five expansion slots and eight drive bays. The expansion slots consist of three PCI slots, one ISA slot and one shared slot, and the drive bays include five RAID-configured bays, two 5.25-inch bays and one 3.5-inch bay. The system includes integrated graphics with 2M of graphics memory, an integrated 10/100 Ethernet controller and embedded dual-channel Ultra/Wide SCSI controllers. Both SCSI ports are open because the server uses a separate RAID controller card, which shipped with 16M of cache that is expandable to 128M. The server contains five 9.1G hot-pluggable IBM Corp.Ultra/Wide 10,000-rpm SCSI hard drives.

Micron's good chassis design features regular screws for securing the case, but all drives except the floppy feature tool-free removal. There is no built-in lock, but you can supply your own to use with the padlock hole included in the case design. However, you cannot lock the front cover, even if you use the padlock hole. We found this worrisome because the RAID array and all buttons are behind the front bezel and therefore cannot be secured. The hard drives are fairly easy to remove without tools by pushing a tab and pulling the drive out. Micron does not offer a tower-to-rack option with this server.

Instead of using proprietary server management software, Micron ships Intel's LANDesk Server Manager with the NetFrame 3100. This package allows you to set thresholds as well as monitor system temperature, fan speed, hardware intrusions and power supply. You also can manage server trends and statistics, which helps in gathering baseline server usage data. In addition, Micron ships a 3Com Corp. modem with the system for remote management, which is a nice touch.

The Mylex Corp. Disk Array Controller utility manages the Micron's RAID configuration. This utility features context-sensitive help, making it more user-friendly than the RAID management software that ships with Gateway's system. The Mylex software lets you fully manage the RAID array and add extra space.

Micron's inclusion of an open standard server management tool and a RAID management utility that features context-sensitive help contributed to its very good management score.

At $9,698, the NetFrame 3100 offers a competitive price. This is a good system with a range of scores that fell in the middle of the pack. With a final score above 7.0, it gets our thumbs-up as a recommended product.

* * *

Gateway's ALR 8200 NTS

Gateway's ALR 8200 NTS landed in third place with a final score of 7.00. Typical for Gateway, this system features excellent expandability - the best of all the systems in this review. It turned in decent performance scores, with a throughput of 1,641 kilobytes/sec and an average response time of 0.22 seconds, but we expected a little more out of it because it is a Gateway ALR server.

The ALR 8200 NTS led the field in expandability, earning an excellent score with eight expansion slots and 13 drive bays. It features six PCI slots, one ISA slot and one shared slot. It is the only server in our comparison to include six RAID-configured bays. This is a nice feature because you can use the sixth drive as a hot spare without losing any throughput performance, as you would if you used one drive as a hot spare in a five-disk array. It also contains five 5.25-inch bays, one external 3.5-inch bay and one internal 3.5-inch bay. The system includes integrated graphics with 2M of graphics memory, an integrated 10/100 Ethernet controller and dual-channel PCI Ultra2 SCSI controllers with both ports left open. The RAID controller card features 16M of cache, which is expandable to 64M. The ALR 8200 NTS is the only system in our review to feature two power-supply bays, and at 400 watts, they are also the most powerful. In addition, they are the only hot-swappable power supplies in this review. The system's six hard drives are 9.1G hot-pluggable, Ultra2 SCSI LVD drives in Seagate Technology Inc.'s 10,000-rpm Cheetah series.

With all this expandability, this system can grow along with your agency, providing good investment protection.

The system's very good chassis design features thumbscrews to remove the side panel, although the drive bays are still the screw type. The tall, thin tower configuration leaves a lot of working room inside, and cables are color-coded, neatly bundled and clearly marked. We liked this unit's security features, which include built-in locks for the front covers (two doors make up the front bezel) and side panel. All buttons are behind the front covers, and all peripherals are inaccessible when the covers are locked. The inside of the cover features a small chart of the motherboard, the memory, the CPU settings and other information, although the data is not as extensive as the Compaq's inside-cover instructions. The hard drives pop out easily without tools. There is no rack-mount kit for this machine, but rack versions are available.

Gateway's proprietary server management software, InforManager, allows you to monitor temperature, voltage, fans, power supply status, ECC memory status and other system information, such as processor and drive activity. You also can monitor hardware intrusions, and the system supports full remote monitoring under Windows NT and Novell Inc. operating systems. The ALR 8200 NTS supports other standard management software, such as Intel's LANDesk, but InforManager provides a quick-and-easy way to check on the system, it is simple to install, and it is intuitive. The one shortcoming we found was the inability to set system thresholds. While we liked the InforManager software, the fact that Gateway did not include a more robust management solution to back it up hurt the product's score slightly.

The ALR 8200 NTS uses RAID management software called ADAC RAID Controller. Accessing drives in the RAID configuration and adding drives to the RAID subsystem was relatively simple for an experienced user, but the DOS-like interface provides no context-sensitive help with making RAID configuration choices. We gave the Gateway product a good management score.

At $10,374, the ALR 8200 NTS is the most expensive system in this roundup, but it is also the only system to feature six hard drives. The unit is an especially good deal when you consider its vast expandability, which translates into investment protection.

* * *

DTK's APRI-81S/P450

Rounding out our comparison in fourth place with a final score of 6.12 is DTK's APRI-81S/P450. This system has unique, innovative features that impressed us, but DTK does not include server management software or truly manageable server-class hardware with the unit, which seriously hurt its score despite the deep-discount price of $5,480. Poor documentation and unacceptable technical support also contributed to the low score.

The APRI-81S/P450 turned in performance scores that were on par with the other systems, with a throughput of 1,968 kilobytes/sec - higher than the Gateway product - and an average response time of 0.22 seconds, which tied the Gateway score. What's remarkable about these scores is that DTK uses IDE hard drives, which are a third of the cost of SCSI drives. DTK wanted to prove that you do not need expensive SCSI drives to turn out good performance, and our test results support this assertion. With strong performance at a low price, DTK might have blown away the competition if the company had included real server management software and hardware as well as better documentation and technical support.

We liked this system's expandability and awarded it a good score for its seven expansion slots and 11 drive bays. The expansion slots include three PCI slots, one shared slot, two ISA slots and one AGP slot. Five of the drive bays are RAID-

configured and are packaged in a third-party add-on from Consensys Corp. called RAIDZone. There are six 5.25-inch bays in addition to the RAIDZone, which leaves enough room to add a second RAIDZone array. The APRI-81S/P450 is the only server in our comparison to feature a graphics card instead of integrated graphics. We thought the Matrox card in the AGP slot was a bit of overkill for a server, which typically does not require much graphics capability. This also is the only unit in our roundup that does not feature an integrated Ethernet controller. The system features built-in Ultra/Wide SCSI controllers with ports integrated onto the motherboard. The SCSI ports are free because the system uses a separate RAID controller. Another unusual aspect of the DTK is that rather than locating cache memory on the RAID controller itself, the system uses software that allows access to the server memory for use as cache. This flexible setup allows users to set the amount of memory to be used as cache for the controller card, with 23 percent of the physical server SDRAM being the maximum. The RAIDZone array contains five 10.1G IBM 7,200-rpm IDE hard drives that, oddly, feature pieces of cardboard between each of them.

DTK's innovation is reflected in its very good chassis design. The system comes on wheels that are lockable and removable. It is rack-mountable by simply removing the wheels and using the existing case, which already has screw holes. We liked the way the lockable front bezel can open in either direction, offering versatility for server placement. All buttons are behind the front cover. Even with the front cover locked, the side panels are removable, so you would have to provide your own lock to secure them. The case cover uses regular screws, as do the non-RAIDZone drive bays. But the RAIDZone array itself and its drives are tool-free.

The lack of included server management software was the crushing blow for the DTK product, resulting in a poor management score. A server without management software is unacceptable for agencies with critical data to manage and store.

The APRI-81S/P450 did include the RAIDZone's third-party management software and manual, which raised the score to poor instead of unacceptable. This software, featuring DOS and Windows NT versions, was intuitive and easy to use.

The DTK's bargain-basement price of $5,480 was at least $4,000 less than the other vendors' products. We liked many things about this server, including the price (the result of using less-expensive IDE drives) and competitive performance scores. It's a shame that the management, documentation and technical support scores were so low. If you want to save a lot of money on your server purchase and you do not mind thin documentation, difficulty reaching technical support staff and lack of hardware management, consider this system, which also offers good overall RAID 5 protection.

* * *

Conclusion

When buying a workgroup server, consider the investment protection afforded by dual Pentium II processors. They enable you to expand your local-area network because they pack plenty of power to run applications, and they have lots of storage space. All the servers in this roundup were decent performers, and all except the DTK fell in the same price range. The scores generally varied the most in chassis design and management. Make sure you examine those aspects before you buy.

-- Pettirossi was the project leader for the FCW Test Center.

* * * * *

AT A GLANCE

Dual Pentium II Workgroup Servers

Pricing: Prices range from $5,480 to $10,374.

What's Selling: Workgroup servers with dual 450 MHz Pentium II processors and robust management tools that monitor every aspect of the system and allow remote management.

What to Specify: Look for systems with at least 256M of memory, full management capabilities, a CD-ROM, dual 450 MHz Pentium II processors and 20G to 45G of RAID-configured hard drive storage and expandable cache.

* * * * *

PRODUCTS TESTED

Compaq Computer Corp.ProLiant 1600Available on the GSA schedule.Score: 8.15

DTK Computer Inc.APRI-81S/P450Available on the GSA schedule, the NIH ECS II contract and the State Department's State Information Infrastructure contract. Score: 6.12

Gateway Inc.ALR 8200 NTSAvailable on the GSA schedule.Score: 7.00

Micron Electronics Inc.NetFrame 3100Available on the GSA schedule and the NTOPS blanket purchase agreement through Pulsar Data Systems.Score: 7.14

* * * * *

Product Features

Compaq ProLiant 1600

Six expansion slots; nine drive bays; a 325-watt power supply; integrated graphics with 1M of graphics memory; an embedded dual-channel integrated Ultra/Wide SCSI controller; a maximum 6M cache on the RAID controller card; five 9.1G hot-pluggable Ultra/Wide 10,000-rpm hard drives; no USB ports; two serial ports; one parallel port; one mouse port; one keyboard port; one external Ultra/Wide SCSI port.

**

DTK APRI-81S/P450

Seven expansion slots; 11 drive bays; a 250-watt power supply; a Matrox graphics card; an Ultra/Wide SCSI controller and ports integrated onto the motherboard; no memory on the RAID controller, but the system uses software that allows access to the server memory for use as cache; five 10.1G IBM 7,200-rpm IDE hard drives; two USB ports; two serial ports; one parallel port; one mouse port; one keyboard port.

**

Gateway ALR 8200 NTS

Eight expansion slots; 13 drive bays; a 400-watt hot-swappable power supply featuring two bays; integrated graphics with 2M of graphics memory; dual-channel PCI Ultra2 SCSI controllers; 16M cache on the RAID controller card, expandable to 64M; six 9.1G hot-pluggable Seagate Cheetah Ultra2 LVD SCSI 10,000-rpm hard drives; two USB ports; two serial ports; one parallel port; one mouse port; one keyboard port.

**

Micron NetFrame 3100

Five expansion slots; eight drive bays; a 300-watt power supply; integrated graphics with 2M of graphics memory; embedded dual-channel Ultra/Wide SCSI controllers; 32M cache on the RAID controller card, expandable to 128M; five 9.1G hot-pluggable IBM Ultra/Wide 10,000-rpm SCSI hard drives. No USB ports. Two serial ports; one parallel port; one mouse port; one keyboard port.

* * * * *

Bluecurve benchmark measures server performance

Benchmark Results

We used Bluecurve Inc.'s Dynameasure/File Professional Edition 2.0, a Windows NT capacity and reliability management tool, to measure the performance of the servers in this review. Results are presented in chart form for the workload discussions. Dynameasure reports on load, throughput, response time and resource utilization. Dynameasure tests are normally run as a series of steps, with an increasing load at each step. Typically, six steps are used, and that process was followed here.

Load

Load is reported in terms of Dynameasure Motors Per Step. A motor runs on a test client machine and represents a user. Motors can be added to apply additional stress to the network. The load applied in each step, in terms of number of motors participating, is indicated on the horizontal axis of each throughput chart.

Throughput

Throughput is the aggregate measure of the work accomplished in the system for all motors, averaged across the number of seconds the workload was applied. Throughput can be reported in two types of units, as either transactions per second or kilobytes per second. Dynameasure reports transactions/sec for all workload types. Kilobytes/sec also is reported for File and Messaging workloads. Throughput charts in this report will use transactions/sec for SQL OLTP and Messaging, and kilobytes/sec for File. In throughput measures, higher numbers are better.

Response Time

Dynameasure reports the average response time per transaction executed, with response time measured in seconds. Lower response times are better. The response time is a measure of the client experience.

Resource Utilization

Dynameasure captures Microsoft Corp. Windows NT and Windows 95 performance counters for several objects, including Logical Disk, CPU and Network Segment. CPU utilization, average disk queue length and network segment utilization are used in this report. CPU utilization is reported as a percentage; 100 percent is a completely busy CPU. Average disk queue length is reported as the count of disk requests waiting plus any request currently being serviced.

This is queue length of the Logical Disk object, so it refers to the queue length for a volume, which may span multiple physical drives. A queue length of two is considered busy for a single drive, so a queue length of 2X, when a volume contains X drives, might indicate a busy volume.

But controllers, RAID levels and other factors influence this as well. Network utilization is reported as a percentage; 100 percent would be a completely busy network. Different network types work differently, however, and some may be completely busy at an indicated utilization below 100 percent. Resource utilization information helps in determining the bottleneck - the limiting resource - under different workloads.

Charts

Charts are used to present results. Each workload discussion usually includes a throughput chart, where the performance curve for each configuration tested is plotted as transactions/sec per motor load level. A higher score is better. Each curve on the chart has a corresponding configuration, indicated in the chart legend. Utilization charts are listed as a percentage of resources utilized for each CPU and the network. For disk resources, the chart represents the average disk queue length. All utilization charts are graphed across time. Dynameasure captured five readings during the measure phase of each step.

Benchmark Results

The Micron server finished on top - about 18 percent faster than the DTK server. Judging from the utilization data, apparently all the configurations were I/O-bound. That is, the limiting factor for each server was the disk subsystem. CPU utilization for each server was low because of insufficient disk capacity to adequately "feed" the CPUs. By improving disk subsystems, we likely would see better server performance in every server.

We'll use the Compaq resource utilization to explain how to read the utilization chart. CPU(0) and CPU(1) represent how busy the server CPUs were during the test. The vertical axis marks how busy the CPUs were. The higher the lines go, the busier the servers were.

Note that CPU(0) never gave more than about 18 percent, and CPU(1) never gave more than about 22 percent. So putting more CPU capacity in the server would not give faster performance. In all servers, CPU(1) consistently had more utilization than CPU(0).

The disk queue length, which can be read on the right-hand vertical axis, refers to the queue length for a volume, which may span multiple physical drives. A queue length of two is considered busy for a single drive. The Compaq, with five drives, appears to approach a bottleneck at around 30 motors, but it is more likely to reach a bottleneck at 45.

* * * * *

How We Tested: Dual 450 MHz PII Servers

We evaluated dual 450 MHz Pentium II workgroup file servers using tests designed to show their performance and feature differences. The test consisted of 40 Pentium and Pentium II clients running over a 100 megabits/sec Ethernet local-area network. Each of the 40 clients ran multiple motors simulating real-world tasks and traffic over the network.

Benchmarks Used

We used Bluecurve's Dynameasure/File, Professional Edition 2.0, which is a Windows NT capacity and reliability management tool, to measure the performance of the servers in this review. Dynameasure employs a close approximation of users performing real work on networked clients and servers. Its components include a test dataset and multiple test scenarios. The test dataset contains the Dynameasure test schema and data. Dynameasure provides a scalable test dataset. This dataset is designed around a schema typical for a given target service, including file services that were utilized in the test performed for this project.

The 2G Dynameasure Standard File dataset is a collection of text, data and image files. The files reside in a shared directory on the target server. (There is no Dynameasure software component on the target server, only data.) The test scenario selected for this project was the Standard File test. A transaction in this test is a file copy, either from client to server or server to client. Each file is 64K in size. Files are read and written in 16K blocks. The files are of six types, including text, image, data and compressed versions of each of those.

We defined the rate at which input/output stress was increased by setting the number of simulated users—called motors—to do the work, adding more motors in successive steps. We increased the stress at a rate of 15 motors per step, starting at 15 motors and ending with 90 at the peak of the test. Dynameasure lets you set the delay, or "thinktime," in seconds between each file copy that the motors perform. In other words, telling Dynameasure to use shorter thinktime is equivalent to telling users to increase their work pace by resting less. After some experimentation, we determined that a thinktime parameter of 1 second created the right amount of stress to bring the servers to the point of diminishing performance; at that point, the number of bytes per second that the servers could process started to decline.

All test resources and activities—client and server selection, dataset generation, and test and result management—are performed from a client running the Dynameasure Manager. Tests queued up and executed from the Manager call upon networked client machines running Dynameasure motors to copy files to and from the target server. During a test, stress is increased in a graduated fashion by adding more motors successively in timed increments, called steps. When a test was finished, results were viewed graphically within Dynameasure or exported to Microsoft Excel. Performance measurements include transactions per second, average response time and utilization (server, client and network).

More information about Bluecurve and Dynameasure is available at www.bluecurve.com.

Benchmark Performance

We scored benchmark performance in two categories: throughput and average response time. We scored throughput mathematically. The system with the greatest throughput received the maximum number of points possible (175). All other units received a percentage of the maximum points based on their slower performance. The average response time also was scored mathematically. The system with the lowest average response time received the maximum number of points possible (75). All other units received a percentage of the maximum points based on their slower performance.

Expandability

We considered the amount of free drive bays in the system, the total memory expansion options, the total number and type of expansion slots in the system, the total hard drive expandability in the system and the overall total expandability after factory configuration. We used word scores that were translated into percentages of the 75 points assigned to this category.

Chassis Design

First, we evaluated overall case design, including how easy it was to open the case and whether screws or other pieces of hardware needed to be removed. We also considered how easy it was to access and manipulate internal components. Quick snap-out drive bays and other special hardware components earned extra points. We also awarded points for system hardware security such as case locks or keyboard locks. Second, we evaluated system ports and features. We awarded points based on how easy it was to replace drives in the array and how well protected the drives were in the case. We then looked at how well power switches were protected from accidental manipulation. Last, we awarded extra points based on new and improved case design features. We used word scores that were translated into percentages of the 100 points assigned to this category.

System Management

We evaluated the server management software that the vendors shipped with the servers. We also looked at features such as thermal monitoring, intrusion detection, CPU temperature, threshold setting and other management features. We installed the management application on the servers and set up a client workstation for remote administration. We evaluated how easy the software was to install and configure. We looked at Desktop Management Interface support, setting alarms, types of notification media—such as paging or e-mail—asset information and integration with third-party Simple Network Management Protocol enterprise management applications. We also evaluated the Redundant Array of Independent Disks software that came with each system, considering how easy it was to use the software and to configure RAID drives in the server. We also looked for any additional monitoring of the drives in the array. We used word scores that were translated into percentages of the 100 points assigned to this category.

Compatibility

We added points for each operating system that would run on each server. We used word scores that were translated into percentages of the 75 points assigned to this category

Documentation

At a minimum, documentation had to tell us how to set up and use the system and had to include accurate diagrams to illustrate the text. Comprehensive, well-organized and well-written manuals received higher scores. We lowered the score if a manual was poorly organized, lacked a table of contents and index, did not include information on installing options or contained factual errors in the text. Certain criteria automatically triggered lower scores; for example, a missing system manual was unacceptable, and a missing software manual earned a poor score. We also looked at any online documentation and judged it on the same criteria listed above. The scores then were added together to come up with a final score. We used word scores that were translated into percentages of the 50 points assigned to this category.

Support Policies

A one-year warranty covering parts, labor and unlimited technical support from the vendor earned a satisfactory score. We awarded bonuses for unconditional money-back guarantees, on-site service included in the purchase price, extended support hours, bulletin board support, such as CompuServe, and a toll-free number. We subtracted points for no technical support, a limited support period and dealer-only support. We then assigned word scores that were translated into percentages of the 100 points assigned to this category.

Technical Support

We based technical support scores on the quality of service we received during multiple anonymous support calls. We also looked at how long we were on hold for each call, and we factored that wait into the overall score for technical support. We assigned word scores that were translated into percentages of the 75 points assigned to this category.

Price

Price was scored mathematically. The lowest-priced unit received the total possible points (175). All other systems received a percentage of the total points.