(See related interview with Vern Brownell, former CTO of Goldman Sachs.)
The name of the game in data centers today is density. With
server proliferation a constant problem, information
technology departments are always on the hunt for ways to
pack more computing power into a smaller space.
They may have found it, with the latest approach to servers,
now beginning to emerge from computer vendors: blade servers.
The standard six-foot high server racks that fill computer
rooms today can hold up to 42 1U servers, each 1-3/4" tall.
But those racks also have 42 power supplies, 42 fans and 42
network cards and cables. By sharing these components among
servers, and removing the sheet metal chassis that surrounds
each server, blade servers squeeze many more computers in
the same amount of space.
The result is a sort of server farm in a rack. Blade server
vendor RLX Technologies, of The Woodlands, Tex., for
example, packs up to 336 servers in a standard 42U rack.
The blade approach emerged in the mid-1990s in the
telecommunications industry, where switching equipment was
mounted in racks using a blade-type standard called Compact
PCI.
Blade servers are only just beginning to appear in
enterprise data centers, however. RLX was founded in
December 1999; another leading vendor, Egenera Inc., of
Marlboro, Mass., opened its doors three months later. Both
of those firms are currently shipping blade-based systems,
as is HP. Sun, IBM and Dell all have announced plans to
introduce blade servers later this year.
So far blade servers seem to be largely popular among ISPs
and ASPs, and for applications such as email, Web and domain
name serving. As they gain acceptance, and as vendors
release more powerful systems, they are likely to start
seeing duty for more mission-critical applications.
Already, Egenera's BladeFrame system, which is targeted at
high-end data centers, is finding its way into computer
rooms on Wall Street. Brokerage house Credit Suisse First
Boston earlier this year replaced 20 traditional RISC-based
servers with a BladeFrame, which it uses to process more
than 60 million financial transactions a day.
Cable Misery
Blade servers pack more processing power into a small
space. They offer another advantage as well, one that at
first glance might not seem tremendously significant: far
fewer cables.
"Cable management is the bane of many a data center
manager's life," says Jonathan Eunice, principle analyst at
Illuminata. Look at the front of a rack of thin servers, he
says, and you see a sleek, elegant computing design. Viewed
from the back, however, the rack is a tangled mess of power,
network, disk storage, keyboard and video cables.
In a large data center, says Eunice, dealing with the tangle
can rapidly get expensive. Today, he says, "you can no
longer afford to have your highly paid IT staff pulling
cables all day, or drawing beautiful maps of what cables go
where."
Because blade servers have internal, high-speed
communication channels between the individual blades, they
can dispense with network cards and cables for each server,
and shared power supplies cut the number of electrical
cords. The result is far fewer cables: Egenera's
BladeFrame, for example, consolidates the cables for up to
24 four-way servers into as few as four cables.
That greatly simplifies administration. "In a traditional
rack," says Sally Stevens, a product manager for blade
servers at HP, "if you want to swap out a server, you have
to go to the back, uncable it, take it out, migrate cards
and recable it. With the blade, you pull out the old one
and snap in a new one. It's as easy as changing a light
bulb."
HP has simplified the process even further, says Stevens,
with the addition of a simple blue light on the front of
each blade. When a blade is shut down to be replaced, the
light comes on, allowing it to be located quickly. That's a
useful feature when a single rack can hold up to 280 blades,
as HP's do.
Automatic Reprovisioning
While blade servers' simple physical architecture is
appealing, the real payoff comes when that is combined with
systems management software specifically designed for them.
HP, Egenera and RLX all offer software packages to manage
blade servers. The software keeps track of how each
particular blade is configured. Replace a blade, "and the
system automatically reprovisions it with the correct
operating system and applications," says Stevens.
Managing the blades through software means companies can
react quickly to changes in the business environment, says
Bob Van Steenberg, RLX's chief technology officer. "As your
computing needs change," he says, "you can reprovision the
blades, entirely through software, to meet those changing
requirements."
It also can dramatically reduce the cost of administering a
data center. "RLX has several customers who've compared
the cost of running traditional computers with the cost of
delivering the same computing capability with blade
computers," says Van Steenberg. "What they've typically
seen is a thirty percent reduction in the cost of operation."
Images
