A promise half-fulfilled
|
Large streams from little fountains flow, Tall
oaks from little acorns grow
- David Everett (1769-1813)
|
Blade
servers seem to have everything right. Theyre modular. You can squeeze
more of them onto a standard rack-mountable blade enclosure than you could fit
conventional rack-mount servers into in the same space. Plus they offer loads
of flexibility in terms of deployment and management. So why isnt everybody
buying blades?
Theres more than one reason why blades are still more
of a curiosity than the de facto form factor for servers in a data centre.
First off, while blades are modular and all that jazz, the cost of buying an
enclosure makes buying one or two blades a no-no. Which is unfortunate as thats
where a small organisation would like to start before it builds out its infrastructure.
The solution for that would have to be the vendor subsidising the cost of the
enclosure and making its money on sales of additional blades. Even if youre
buying a fully populated enclosure, blades cost a bit more than traditional
rack-mounts though that gap is closing.
Beyond that, if we are going to talk about standards, why cant you just
plug in a blade from one vendor into an enclosure made by a second vendor. That
would be a compelling reason to choose blades over rack-mounts which can be
fitted into the same rack regardless of their manufacturer. Without this capability
youre locked into a relationship with a vendor for expanding your blade
deployment giving you little flexibility if your vendor decides to go slow or
forego making, say, Opteron-based blades or 2-way blades or even RISC blades.
Blades are a promising architecture which is why server and networking vendors
have both started building them. If true interoperability came about you would
end up in a world where every switch could be a server and vice versa. Then
there are storage blades and unclassifiable ones like IBMs Cell processor-based
model that has nine dual cores and is aimed at medical imaging and life sciences
applications.
While theres always going to be a need for 32-or 64-way
SMP monsters to run mission-critical applications, for the restblades
will do very nicely once the pricing and vendor interoperability issues are
sorted out.
Prashant L Rao
Head of Editorial Operations
|