By Yuval Shavit, Features Writer
With all the hype around them, it's a fair question to ask what blade servers can do that standard rack-mounted servers can't. To a certain degree, the answer is: not much. A blade is basically a standard server stripped of its noncomputational parts, such as power converters and direct-attached storage (DAS), so its functionality is nearly identical to any other computer. But using blade servers for advanced IT functions, like server virtualization or high-performance database clusters, is increasingly common.
One of the original uses for blade servers was the "one server, one function" paradigm, in which blades could serve as slimmed-down versions of standard servers, said Eric Nelson, director of business development at Alteritech Inc., a systems integrator (SI) and managed service provider (MSP) in Vienna, Va. But because blade chassis are expensive, blade servers have an overhead that only balances out if you buy enough to fill each chassis you buy, Nelson said. And since consolidation is a big trend in IT, blades these days are more often used in conjunction with server virtualization.
The idea behind server virtualization is to run several servers -- including their OS, configurations and applications -- on one physical computer. Because computers have a significant amount of CPU overhead, one server running at 80% utilization will be able to handle more tasks than four comparable servers running at 20% each, said Dan Olds, principal at Gabriel Consulting Group Inc. in Beaverton, Ore. Virtualization can therefore save your client a significant amount in energy costs. Many companies are choosing to implement that virtualization on blade servers for efficiencies of shared components that blade servers bring.
Depending on how many servers your client needs, blade servers may or may not make sense. Your client may not have many servers to begin with, in which case consolidating them may require fewer units than would fill a blade chassis. In this case, you're better off suggesting virtualization on standard servers. If your client has enough servers to justify blades, they may save server room space and money in energy costs over the long run. On the other hand, blades are harder to cool, and the extra cooling equipment they require can sometimes eat into some or all of the savings for money and server room space. (We will take a closer look at advanced cooling techniques in the next installment of this tutorial.)
Whether or not you use blade servers for virtualization, make sure to buy a few more blade servers than the minimum. Virtualization puts more of your client's eggs in each basket, so it's important to have a redundant, high-availability system so that your client doesn't suffer a multi-server crash if one piece of hardware fails, said Scott Gordon, sales engineer at ActivSupport, a networking consultancy in San Bruno, Calif.
Virtualized servers don't perform well under intense I/O operations, so database servers aren't generally good candidates for virtualization. On the other hand, modern database applications support clustering, such as with Oracle RAC or Microsoft's federated databases, which is another of the uses for blade servers. Clustering is known as scaling out a database, as opposed to scaling it up, by beefing up a single server, and it works well with blades' density.
Clustered databases need shared storage, which blades also require, so together they can make a compelling argument for a new storage area network (SAN) system. Although the cost of a SAN used to be prohibitive for many SMBs, technology like iSCSI has brought the price of a SAN down. You can also set up network-attached storage (NAS), although that may not perform fast enough for the kind of high-availability databases that a clustered set of blades would provide.