Blade Servers in the Enterprise
More on blade servers
The market for blade servers has finally caught up with the buzz that surrounds them. Blades are one way to cram more into less space, and they also allow for more effective hardware management. This article examines the types of features are available and touches on the difference between what the Big Four vendors in this space offer.
Blade Servers, DescribedWhat makes blade servers be considered blade servers, and which features make for a sharp offering?
What is a blade server, you ask? A standard rack can fit 42U (units) of space, which means you can install 42 1U servers per rack. Assuming you have the power and cooling capacity, that is. A blade server starts with a chassis, which normally provides power and cooling for all the servers that will be inserted. Which components live on which part (chassis vs. each blade) is dependent on the manufacturer and wildly varies.
For example, if a 42U rack can house six 7U chassis that each contain 14 slots for servers, an 82-server density is possible. Some configurations, especially those with taller racks, allow for more than 100 servers each. The capacity increase is enough to persuade many enterprises to buy a blade server, but there are other important benefits to converting.
Depending on the manufacturer, many features accommodate ease of management. Most blade servers integrate the standard server components: networking, KVM, and power and cooling. Determining what level of integration actually was implemented is a per-server and per-vendor challenge. In the networking department, most blade servers have optional gigabit and Fibre Channel switch modules. Only one (or more for redundancy or throughput) connection is required per-chassis if the particular chassis supports networking. Similarly, an entire blade server consisting of 14 servers can connect to a SAN with only two fabric ports being used.
On the power and cooling front, many blade servers use two 2000W (or higher) power supplies. Many of these require 3-phase 240V power, which is an implementation consideration for data centers that don't have this type of power. The advantage is that 3-phase power operates much more efficiently. You can power a full chassis with much less power than is required to run 14 1U servers. The cooling is also centralized per-chassis, so less power is consumed turning high-speed fans within each server.
Although no longer the undisputed leader in the server blade market, IBM still commands a strong presence. IBM BladeCenter servers are available in three varieties. For all three, IBM's approach is to implement power and cooling in the chassis. Each server is a distinct real-life server, complete with CPU, RAM and hard drives. With the optional Fibre Channel or gigabit network blades, network infrastructure can be consolidated. IBM offers two different blade chassis that are quite similar. The standard BladeCenter allows 14 blades, which can be filled with mixed-and-matched IBM POWER or AMD Opteron servers.
Where the two similar BladeCenter servers differ is their target markets. One is aimed squarely at the midrange; the other, called the BladeCenter H Chassis, is geared toward enterprise operations. The H provides beefier power supplies and adds Infiniband capabilities. Both are suitable for high-performance computing, business applications and database applications. The third option, the BladeCenter T-series chassis, is specifically designed for telecommunications applications.
HP shares the top spot with IBM in the blade space. Its BladeSystem server options cover a variety of HP ProLiant servers and are available with both Xeon and Opteron processors. In July, HP introduced its C-class blade infrastructure as the successor to its p-class line.
The older p-class line offers several power options (e.g., a 1U and 3U additional chassis that enable customers to select their desired capacity and optional Cisco gigabit Ethernet card that provides real switch functionality on-chassis). HP plans to continue selling the p-Class through 2007, and it will offer service and support through 2012.
HP boasts that compared to the p-Class, its BladeSystem c-Class blades are simpler to manage, consume less energy and have more bandwidth in the backplane. Result: Each 17-inch box can hold as many as 16 servers or storage devices.
c-Class blade options include the latest and greatest in processors, as well. In August, the OEM announced a 2-socket blade and a 4-socket blade available with the AMD Opteron Rev F. processors.
Dell is a recent entrant on the current blade landscape. Its PowerEdge 1955 is a blade chassis that accepts 10 servers in a 7U footprint. Each blade can be a 2-processor Intel Xeon server, with up to 146GB of storage. Dell's Web site lists a host of market focuses, each with a “1” next to it. Selecting e-business, database, high-performance computing or Web services all lead to the 1955 server, indicating Dell claims the server is suitable for all applications.
Egenera, however, offers a completely different kettle of fish. Both its ES and EX line of blade servers provide a virtualized computing environment. The ES is designed for more light-weight applications, while the EX is ready to take on the most daunting computational loads. Of course, networking and storage (SAN) access is handled in the same way.
Each blade in the Egenera chassis becomes a pool of resources that every operating system can access. The pool is called a Processing Area Network (PAN), and the operating system isn't aware it exists. The control blades virtualize everything, allowing multiple operating systems to run concurrently and share resources in the PAN. Since the processors themselves are virtualized, you can mix and match Opteron and Xeon boards, and pull them at any time for maintenance while the system is running. Linux, Windows, and Solaris 10 are all supported.
Why Blade Servers, Anyway?
Total cost of ownership, return on investment, yada yada yada. But the reality is that blade servers really do save tremendous amounts of space, power and administrative time dealing with hardware. In the case of the traditional blade server, where one operating system runs on each blade, not much is saved in terms of configuration maintenance because the same number of operating systems must be tended to. The management advantage is realized when the same number of administrators can easily handle twice as many servers as before because less time is spent dealing with hardware issues.
As virtualization of all types becomes more prevalent, hopefully more vendors will start making blade servers that provide a virtual server environment. The additional flexibility gained from a virtualized architecture enables IT organizations to approach the marketing claims of traditional blade systems.
Drew Robb is a freelance writer specializing in technology and engineering. Currently living in California, he is originally from Scotland, where he received a degree in geology and geography from the University of Strathclyde. He is the author of Server Disk Management in a Windows Environment (CRC Press).