Blade Servers

Just 30 years ago, a high-end computer easily filled an entire room. But such "big iron" systems are now the exception. Most modern data centers consist of racks of servers. Over time, systems have shrunk to the point at which a rack can easily hold a dozen servers, but the need for more power in less space continues.

Blade servers are the next advancement in the ever-shrinking hardware footprint. They look similar to PCs but are designed for mounting in a rack. A few years ago, a rack-mounted server commonly required two or three mounting bracket units—2U (3.5") or 3U (5.25")—of cabinet space. More recently, manufacturers have started introducing systems that require only 1U (1.75") of cabinet space. Suddenly a rack that was limited to holding nine or ten 2U servers can house twenty servers.

Reducing the server footprint eases systems management because it lets data centers expand their capabilities without increasing the size of their facilities. The latest improvement is known as the Web blade. Web blades fit into a blade server and provide a greater density of servers in less space and lessen maintenance costs by reducing the number of components that can fail. Because Web blades operate on the basis of scaling out, they're ideal for load balancing Web sites but aren't well suited for, say, clustering solutions. However, manufacturers are making the move to combine the scalability of Web blade-server architecture with fault-tolerant server architecture to maximize reliability and scalability.

Web Blades
Although the term Web blade is relatively new, some companies such as Cubix have been designing blade servers for years. Instead of mounting individual systems horizontally, as in a conventional server rack, you mount each Web blade vertically in a blade server. In essence, the blade server provides a common backplane for the Web blades to snap into. In addition to enabling increased density, reduced cabling, and shared power, a blade server uses one set of baseline components (e.g., CD-ROM or DVD drive, power supply, network cards) to support multiple Web blades. These shared resources help to reduce maintenance costs and power consumption.

The blade-server market is growing fast, and several new entries, including Dell's PowerEdge 1655MC, the IBM eServer BladeCenter, and Hewlett-Packard's (HP's) ProLiant BL20p G2, have emerged during the past year. However, not all players in this market are new. For example, Cubix has been working in the blade-server market for almost 10 years. The company's latest product, the BladeStation, squeezes up to seven Intel Xeon dual-processor Web blades into 6U (10.5") of rack space. The design supports four hot-pluggable SCSI drives on each Web blade. The design is supported by a scalable array of power supplies that reduce power requirements over a series of 1U servers and provide redundant power across the systems. A Web blade configuration costs about the same as a standard server setup; the BladeStation starts at less than $3000.

Although blade-server manufacturers often work on the leading edge of technology, some of their blade-server rack designs are compatible only with blade servers from the same manufacturers, leading to the perception in the IT industry that blade servers are proprietary. To overcome this perception, blade-server manufacturers are starting to ensure that their systems contain open architectures. For example, Cubix's XP4 blade server includes a standard PCI slot that lets you add a generic acceleration card or a product such as 14 South Networks' SlotServer, which can provide firewall capabilities at the server level.

Another 14 South Networks blade-specific product, the SlotShield PCI card, lets administrators offload firewall and other server security capabilities from a blade-server's processor. For example, in a Web hosting environment or in those instances where you want to assign one blade in a blade server to meet the needs of a particular customer, the SlotShield card lets the server become a standalone solution.

Managing Blades
Most Web blades come with a standard OS such as Windows 2000 Server and often provide integrated NIC capabilities and automatic failure-detection logic to provide basic load balancing. To this basic configuration, you can add deployment and management tools such as Microsoft Application Center 2000.

Part of the value of blade-server architecture is its affinity for load balancing. Being able to group all the blades associated with a Web site into one server offers a simple management solution. Although you can manage each blade individually, a more common method is to employ blade-management software.

Blade servers all ship with some form of management software, but the unique architectures tend to require custom software to manage the components. In many cases, this level of management leaves administrators addressing multiple servers (blades) individually in each blade server. Managing these blades is one of the main challenges when you implement blade servers.

One such load-balancing and availability manager that's customized for blade servers is F5 Networks' BIG-IP Blade Controller software. With a starting price of about $4000, which can easily jump to more than $15,000 to manage a set of Web blades, the management software isn't cheap. You can also add related products from F5 Networks to integrate your blade servers with nonblade systems. Regardless of how well you manage load balancing and failure detection, these management capabilities don't replace the failover and reliability of clustered servers.

Fault-Tolerant Blades
In "Fault-Tolerant Servers," January 2003, http://www.winnetmag.com, InstantDoc ID 27403, Mark Weitz explains that several vendors have developed server products that provide clustering and use parallel execution to provide instantaneous failover. Because these servers manage clustering at the hardware level, they can bypass many software-level delays associated with detecting faults and failing over.

Recently, NEC released an updated version of its Express5800/320Lb fault-tolerant system that's based on Xeon processors. With its latest entry, NEC is making its architecture more compatible with standard blade-server architecture. The Express5800/320Lb makes use of a blade-server architecture that cuts NEC's fault-tolerant-architecture footprint from 8U (14") to 4U (7") and includes support for Intel's Hyper-Threading Technology.

The Express5800/320Lb consists of a pair of fault-tolerant blades, each of which can have one or two physical processors. You can virtually double the number of processors by taking advantage of Hyper-Threading Technology. This technology supports increased performance on Windows servers by letting one high-end CPU create two command streams and function logically as two processors. However, unlike a Web blade, a fault-tolerant architecture requires two types of blades: one for I/O and disk operations and the other for CPU and server-processing elements.

The differences between the two types of blades are necessitated in part by the fact that unlike a traditional Web blade that scales out, the NEC solution is designed to increase reliability through failover. Whereas traditional Web blades co-locate several independent servers, fault-tolerant architectures run co-located server pairs in parallel. As a result, these fault-tolerant servers are best suited for database and email servers that require transactions and can't risk interruption by a failed system. For maximum failover capabilities, you can run cluster software across fault-tolerant server pairs to achieve reliability both within each server and across a traditional cluster.

When a blade fails, a technician simply removes that component and installs a replacement—the system never sees any interruption. Of course, this level of reliability comes at a price. The dual hardware and synchronized execution results in a price tag of almost $30,000.

Choosing the Right Solution
Selecting the right server often comes down to how you plan to use the hardware. If you plan to use blade servers to support Web sites, you're best off selecting a traditional blade architecture as opposed to a fault-tolerant architecture. When you're evaluating your options, look for support for features such as the ability to integrate a PCI card or run a particular version of management software. The key is to examine your hardware requirements, then prioritize the options available to you.