This version of this QuickStudy article originally appeared in Computerworld's print edition.
- Blade servers are modular, single-board computers, typically about 7 in. high, 2 in. wide and 19 in. deep. Each blade contains processors, memory, network controllers and other I/O ports; it plugs into an enclosure that holds multiple blades and provides power, cooling, networking, specialized interconnects and management.
Blade servers were invented to enable today's small, powerful computers to fit more efficiently into standard server racks, whose large physical size was determined back when electronics required vacuum tubes. Server racks are 19 in. wide and are designed to hold components that measure a multiple of 1.75 in. high (a measurement referred to as U). The standard rack is 42U high, so 42 1U-high servers can fit into a rack.
But a 1U server takes up very little of the horizontal space in the server rack. To use that space more efficiently, individual blade servers are mounted vertically in a 6U- or 7U-high blade enclosure, plugging into a backplane that can hold up to 16 servers rather than the six or seven that would fit horizontally.
Technically, blade servers harken back to 1981's VMEbus architecture, which allowed a single-board computer to be plugged into a chassis backplane with multiple slots. In 2001, the PCI Industrial Computer Manufacturers Group adopted a backplane/blade structure in which a single enclosure could include multiple computers with one master board coordinating the entire system.
When a single board included the CPU, memory, I/O and nonvolatile program storage -- in other words, a complete server, with operating system and applications -- it came to be called a blade server. RLX Technologies, a Houston firm made up mostly of ex-Compaq employees, shipped the first blade server in May 2001. RLX was acquired by Hewlett-Packard Co. in 2005.
Blade servers plug into a chassis backplane in an enclosure that can hold six to 16 blades. The enclosure balances electrical power according to the various component blades' demands. A fully loaded blade server system can generate considerable heat, so enclosures monitor it and may shut down the entire system if the temperature rises too much.
Hot swapping -- the ability to add, remove and replace server blades without powering off the system -- is an important capability. A problem blade can be removed and repaired or replaced without disrupting others in the same enclosure.
Blade servers generally connect to the enclosure via Ethernet. Each enclosure has Ethernet and/or Fibre Channel switches connecting each blade server to the LAN. The enclosure also includes USB and VGA ports for monitor, mouse and keyboard I/O connections, and it may contain a CD or DVD drive. Blade servers simplify cabling; as many as 200 cables coming out of a "normal" rack can sometimes be replaced with just three to six cables.
Each blade server may have local disk storage, but using a storage-area network keeps both enclosure and blades completely free of storage systems' inherent heat, noise and reliability problems. Open blade architectures exist, but, in practice, blade servers most often use proprietary connections.
IDC identifies HP, IBM, Sun Microsystems Inc. and Egenera Inc. as the major players in the blade market. The research firm says that server blades were the fastest-growing server segment in the third quarter of 2007, accounting for more than 10% of all servers shipped. HP led the blade market in that quarter with 42.1% market share and factory revenues that were up 79.6% from the same period a year earlier. IBM held the No. 2 spot with 32.9% market share.
Kay is a Computerworld contributing writer in Worcester, Mass. Contact him at firstname.lastname@example.org.
For a complete archive of QuickStudies, go to computerworld.com/quickstudies.
For more information on blade servers, visit these sites: