Blade servers gain momentum

IBM offers internal Fibre Channel switch

                                          By Lisa Coleman

No longer just a niche market, blade servers are quickly gaining popularity and are expected to grow into a $2.5 billion market by 2005, according to IMEX Research, in San Jose.

This year, the blade server market is estimated at about $120 million, according to International Data Corp., and IDC predicts that by 2006, 20% of all servers shipped will be blade servers.

Server blades--independent servers containing processors, memory, network/storage I/O, etc., mounted on single boards, or blades--plug into a chassis, which sits in a rack sharing common components such as power supplies, fans, disk storage, and Ethernet or Fibre Channel switches. Blades evolved from 1U (1.75-inch-high) server designs and are inserted vertically into a chassis. Therefore, hundreds of processors can fit into the space that could previously only hold dozens of processors on 1U server boards.

Blades offer significant advantages to IT managers, allowing them to consolidate resources (servers, storage, and networking) and save money on power, installation, and rack/floor space. Blade proponents also claim that systems management and deployment are far easier due to centralized management and "rip-and-replace" serviceability.

The blade server market emerged in 2001 when some vendors introduced dense blade servers for xSPs. Major vendors such as Dell, Hewlett-Packard, and IBM quickly followed suit.

Blades can attach to internal rack-mounted storage, or to external arrays--including network-attached storage (NAS) servers. Emerging designs allow connection to storage area networks (SANs) via internal switches (see "Blade servers offer new twists for storage," below).

Driving blade server market growth is a need for users to better manage their infrastructure, according to Tim Dougherty, IBM's director of blade marketing and strategy. "We're much more focused on how to build and manage this enterprise infrastructure than we are on the power, cooling, and cabling savings, which are inherent in blade server technology," says Dougherty.

Blade servers offer huge scalability in a small space and flexibility in the type of blades that can be used within a single chassis--from low-cost uniprocessor blades to high-performance multi-processor blades. But the real key is management, according to Bob Van Steenberg, chief technology officer and vice president of platform design at blade vendor RLX Technologies.

Since blades lack video, keyboards, and monitors, users cannot just plug into them for management. Integrated processors monitor blade status, the chassis, storage, and switches.

One key to understanding the lure of blades is their volume-driven economics, explains Anil Vasudeva, principal analyst at IMEX Research. Increased volumes will drive down blade server costs, and by 2005, the "sweet spot" for blade servers will be dual-processor and quad-processor blades, says Vasudeva.

Today, blades come in several types and are essentially "bricks" that can be used to build different architectures. For example, typical edge servers are low-power blades in dense packaging that's less than 1U high. Application servers are symmetrical multi-processing (SMP) blades with lots of memory and multiple blade options, typically 1U high. Embedded or telecom servers range from 1U to 3U and have PCI or Compact PCI blades with Network Equipment Building Standards (NEBS) compliance.

Blade server start-ups initially focused on edge servers or single-function applications such as Web serving or caching. However, most blade server vendors now see blades operating in all tiers of the IT infrastructure.

Vendors are primarily targeting four markets: telecommunications, mid-sized to large enterprises for commercial computing, high-performance and technical computing, and xSPs, according to Vasudeva. Within these markets, specific applications such as online transaction processing and video streaming are driving the availability of different types of blades and the types of storage necessary to run those applications.

Blade servers offer new twists for storage

More than 150 companies are involved in some part of the blade server market, which is expected to grow to $2.5 billion by 2005, according to IMEX Research. Many new entrants are expected next year, including Sun Microsystems. Meanwhile, IBM is introducing a blade server with an integrated Fibre Channel switch.

This year, IBM began shipping its eServer BladeCenter, based on Intel's Xeon processors. Each BladeCenter chassis is 7U high and holds 14 blades. A typical configuration might include six blades for Web serving, three blades for collaboration, two for terminal serving, two for file serving, and one as a spare. A total of 84 blades fit per standard rack.

Next month, IBM will introduce an optional internal Fibre Channel switch (equivalent to a 16-port external switch) for its BladeCenters. The integrated switch (which is manufactured by QLogic) will allow the blades to connect to a storage area network (SAN) fabric. Each blade that uses the switch requires a Fibre Channel daughter card.

"Clearly, SANs are part of the enterprise fabric today, and customers want access to that," says Tim Dougherty, IBM's director of blade marketing and strategy.

IBM's blades can hook into its FAStT storage arrays. This architecture will make SAN infrastructures simpler and less costly to manage, according to Dougherty. By using the blade's integrated switches to implement a Fibre Channel fabric, IBM claims users can save up to 25% compared to typical rack-optimized servers.

Blade management is delivered via IBM Director systems management software, which allows for automatic deployment of applications and blade upgrades. It also provides management and monitoring for the hardware and applications.

In 2001, Hewlett-Packard introduced its Powerbar line of blade servers. Then, Compaq (before the HP merger) introduced its BL series of blades based on Pentium III processors. Now, HP is selling the ProLiant BL p-Class blade servers with Pentium III processors and an integrated RAID controller. Last year, Dell released its first blade server, the PowerEdge 1655MC, with a single Pentium III. (The 1655MC is priced from $1,499.) Meanwhile, Sun will roll out blade servers over the next two quarters to target distributed applications such as Web services, technical computing, and financial applications. Hitachi Data Systems will ship a network-attached storage (NAS) blade later this quarter.

One company that has been shipping blade servers for more than 18 months is Houston, TX-based RLX Technologies, which initially focused on the Web-hosting market. While RLX's ServerBlades are most frequently used to connect to NAS, its customers are starting to use blades to solve scale-out problems.

"Storage problems are kind of like scale-out problems," says Bob Van Steenberg, chief technology officer and vice president of platform design at RLX. "All these storage domains are examples of where you need some computation and lots of storage. If you can spread it out in modular building blocks [blades] and do parallel processing, you can solve a big storage problem in an innovative way."

Some RLX customers are using its blades with small, inexpensive processors and putting as much disk as possible on them, treating them as storage blades for scale-out problems. In addition, they add application software to the blade and treat it as a big storage system, explains Van Steenberg.

"The blade concept may play not only in servers that connect to storage, but in fact may be a new way of building storage systems," says Van Steenberg.

 

Find this article at:
http://is.pennnet.com/Articles/Article_Display.cfm?Section=OnlineArticles&SubSection=Display&PUBLICATION_ID=23&ARTICLE_ID=163321&pc=ENL