There are a lot of appliances in my kitchen: refrigerator, cook top, oven, microwave, dishwasher, toaster, coffeemaker. They all have specialized functions. No one suggests there should be a single, general-purpose kitchen appliance that could be adapted to cover all these tasks. An appliance model—natural for the kitchen—might also be just as natural for the data center.
IT appliances are already there. A router is just a network appliance for sending and receiving packets efficiently. There are also specialized computing platforms for storage, search, security, e-mail management, data analytics and system management. They all have similar objectives: do a few things very well in order to outperform a general-purpose computing architecture that relies on layers of software to perform the same tasks. Think of it as a race between processor cycles applied to specialized tasks, and the same cycles applied to generalized tasks. A general-purpose architecture has to cope with a much wider range of usage scenarios. On the other hand, appliances don’t have to be good at everything, just the functions incorporated in the devices.
The availability of industry-standard components and Linux have radically altered the economics of the appliance approach. Appliances benefit from the established set of monitoring and management standards that let them behave as good citizens in the network in a way that general-purpose computing platforms and software can’t.
Today, most appliances are focused on at least one of three categories:
Think of the appliance model as the anti-virtualization strategy. As general-purpose computing platforms adopt virtualization to get higher levels of utilization for diverse workloads, appliance-makers target workloads that perform best on specialized platforms, and then build dedicated appliances for those workloads.
This is another of those sourcing dilemmas that CIOs often face. It would be nice if everything in an IT system was a software function that executed on industry-standard virtualized hardware pools. But if a company goes that route, it must depend on its software engineers for adequate performance and reliability, and it may never get enough of either. And software is relatively more expensive to acquire and support over an extended lifecycle. Yet, if a company has a data center full of disparate, discrete appliances all focused on specific tasks, it has much more administrative complexity to address. The individual assets may be cheaper, but the skills and capacity needed to use them could erode the cost advantages that led the company down the appliance route in the first place.
Co-existence is the key. IT strategists should be looking hard at appliances when performance or scale is critical, but shouldn’t abandon general-purpose computing platforms just yet.
