Don’t want a footie-field-size data centre? No problem (or is there?)

Different workloads

More fundamentally, though, small businesses simply have different design considerations to hyperscale cloud providers, warns Hill.

“You also have to look at the fact that a lot of the design decisions made by people at Facebook were made by people facing the specific kinds of workloads that they face there,” said Hill. “The typical enterprise, and especially the small business, doesn’t really have the same kind of problems or need the same kinds of solutions that a cloud-scale environment does.”

The likes of Facebook aren’t buying thousands of servers to pack them each with as many different applications as they can. Instead, they want to spread vast workloads across them, having them all run tiny pieces of the same stuff. A small business, on the other hand, wants to try and pack its accounting and marketing apps into virtualized operating systems on the same physical server to save a few bob.

Cloud-service providers may thrive on standard server builds for generic, high-volume server installs, but that doesn’t mean that Open Compute can’t handle a variety of needs, protests Matt Kimball, a senior product marketing manager at AMD. The enterprises gradually adopting the standard prove that, he argued.

“While their workloads may be slightly generic, the OS image is anything but generic. In many instances, these companies have their own OS instances and a variety of add-in cards and management stacks that are not so ‘vanilla’,” he said.

In essence, Open Compute-based servers are no different than servers designed and manufactured by the major OEMs, points out Kimball. The components and processors are usually from the same ODMs, which are the companies that supply the OEMs in the first place. This is going to make it cheaper for small firms to use Open Compute kit, he suggests.

“Not to trivialize the process of deploying Open Compute based servers, but this should not be a difficult process. Build the OS image, clone it, and away you go," said Kimball.

A price comparison

The cost savings can make it worthwhile, he suggests. AMD carried out a price/performance test, comparing an OpenCompute-specced server against an HP DL360p. The team ran the LoginVSI performance measurement system, and DVDStore to test SQL Server performance.

“What we found in both cases is that the Open-Compute box ran similar performance with substantial savings,” he said. “This maps over to what we are seeing in the SMB space, that the number-one factor in purchasing server hardware is price. Not manageability, not top-bin performance, not even security, but price.”

All of the cost savings came in the capital expenditure, the report showed. The OCP box cost $4,589 in up front purchase costs, compared with the HP’s $10,501. There were several reasons for this, one of which was that you’re not paying your slice of the OEM’s for the marketing budget.

To be fair, though, it also stripped out “certain components that are commonly found on commercial OEM servers”. These included the storage IO controller, arguing the OEMs mark these up when they’re placed on the board, and they aren’t always needed in some workloads.

Rackspace’s Sullivan points out that this can reduce expensive ongoing operational costs that might otherwise normally be incurred by users of OEM equipment.

“I think one of the biggest potential benefits for smaller businesses (besides lower cost of equipment) is the absence of licensing costs pertaining to hardware/firmware features on a typical Open Compute motherboard,” Sullivan explained. “Many mainstream, non-OCP offerings charge recurring licensing fees for out-of-band management and other features or functions. These get implemented in the form of licence keys with expiration dates, etc. I have not yet seen an OCP offering that charges for these sorts of things.”