Blades with an edge

BLADES DECONSTRUCT the enterprise server to the bare minimum — processors, memory, chipset, hard drives, and network interfaces — mounting all on a circuit board that’s stuffed into a high-density rack-mount enclosure. The enclosure provides power and management and allows an entire server to act as a hot-swappable, field-replaceable unit.

This means real benefits. Servers consume far less physical space, thereby requiring fewer racks; the servers consume less electricity, saving power and operating costs; and if implemented properly, the servers are also easier to manage than their traditional 1U-high or larger counterparts, both individually and collectively. Plus, swapping out a bad blade is an order of magnitude easier than deracking and reracking a traditional server.

The challenge for blade vendors is to increase density and improve manageability, while packing as much reliability into a blade as possible. That’s where RLX Technologies comes in: The company’s ServerBlade 1200i server, System 300ex enclosure, and Control Tower 4.0 management system are simply the finest executions we’ve seen on the blade premise.

The meat of a blade system is the server. RLX offers a variety of blades (including two models based on Transmeta’s Crusoe processor), but the one we reviewed, the ServerBlade 1200i, is the flagship of the company’s fleet. Think of it as a beefed-up laptop motherboard, without a keyboard or graphics chip but with a 1.2GHz Intel Mobile Pentium III processor and one DIMM slot (for a maximum 2GB of RAM).

Although the 1200i is available diskless, there’s a built-in ATA-100 disk controller and room for two 2.5-inch hard drives (ranging from 20GB to 60GB). Applications can use two Fast Ethernet NICs; a third is provided for management. There’s also a front-mounted serial port for a debug console.

The 1200i is a good, basic server, but far slower than many modern servers that boast one or two multigigahertz Xeon processors. Forget using 1200i blades for data mining, but they are adequate for Web serving, domain controllers, and so on, especially when clustered or load-balanced (see ” F5 balances blade loads “). In many ways, the individual servers are comparable to HP’s ProLiant BL10e blades, except that the RLX blade supports two hard drives, while the HP blade offers a KVM (keyboard, video, mouse switch).

The RLX 1200i blades are most beneficial when running Linux. There, the server’s lack of keyboard, mouse, and monitor would be considered an asset, not a liability. We were sent two blades with Red Hat Linux 7.3; they were fast and reliable, and we were perfectly able to work with them via SSH (Secure Shell). In fact, the entire RLX system was easy to install, configure, and operate.

Although RLX supports Windows 2000 Server and sent us a blade with that operating system configured for remote command-line access, a Windows server is best managed with a graphical console. Windows is also less happy working with a network-based CD-ROM for software installs, and since the RLX servers lack a USB connection or any other way of directly controlling a CD, that’s another strike.

RLX makes several enclosures of different sizes for its blades. We reviewed the largest, called System 300ex, which is a 3U-high (5.25-inch) box with 24 slots, dual hot-swappable power supplies, and an active management back-plane that allows the RLX Control Tower server to identify blades based on their physical location within the enclosure. But it goes one better: Small dip-switches allow you to set a rack number and enclosure number within a rack, allowing an RLX system to uniquely identify each blade in a datacenter, and dynamically integrate it into a management profile. For example, if a blade fails, replacing it with a new one automatically assigns the replacement with the same data image and IP address.

In theory, the System 300ex can hold as many as 24 blades, but that density is only attainable with the 667MHz or 800MHz Transmeta servers (and presumably with the new 1GHz models, as well). The 1.2GHz Intel-based blades are double-width, allowing a maximum of 12 per 3U enclosure, or 168 servers per standard 42U rack. By comparison, HP’s BP10e system allows 20 servers into a 3U rack.

You can mix and match all of RLX’s current blades in the System 300ex enclosure. But RLX plans to introduce more powerful blades later this year, based on Xeon processors, and says that the power and cooling requirements will necessitate a new, and incompatible, physical form-factor.

On the downside, the System 300ex enclosure creates a rat’s nest of cabling. The enclosure back-plane passes as many as 48 Ethernet connections — two per slot — to four 12-port RJ-21 connectors. If you’re using Ethernet switches that support RJ-21 connectors, such as Cisco’s Catalyst 6xxx series, that’s great. If not, RLX provides four cables that convert each RJ-21 connector to 12 RJ-45 cables. That was a nuisance with only one 300ex enclosure, and it would be a nightmare for a fully populated rack. (This year, the company will be offering an option to replace the RJ-21 connectors with Gigabit Ethernet switches, each of which will consolidate 12 Fast Ethernet ports, but has no plans to incorporate Gigabit Ethernet into the blades themselves.)

We have nothing but admiration for Control Tower 4.0, a management system built onto an 800MHz blade running Linux. Through either a browser or SSH interface, the SNMP-based control shows system status and alerts through the entire RLX ecosystem, and also provides the facility for capturing complete disk images from blades and redeploying them to other blades, or using them to automatically restore the blades. Control Tower’s simplicity and tight integration stands in stark contrast to HP’s cumbersome Insight Manager(see ” HP blades sharp and dull “). Although there is no hard limit to the number of servers that can be managed by a Control Tower 4 server, the company advises a practical limit of five or six chassis per Control Tower.

The RLX system demonstrates the right approach for high-density blade systems. There are many places that large numbers of relatively small servers could be beneficially employed, including hosting environments, schools, test labs, and Web farms. Of course, given the specs of the individual blades, and their individual lack of expandability, there are places they don’t fit as well, such as on storage-area networks, number-crunching, and the database tier of an enterprise-scale application. But if the task calls for a blade, RLX has honed it close to perfection.