Say what you will about a diminished Microsoft in an increasingly cloud universe, the company still has the ability to shake things up from time to time.
Redmond did it again this week with the announcement that it is contributing its custom server design and related technologies to the Open Compute Project, the Facebook-led effort to streamline and commoditize data center infrastructure for hyperscale deployments. The move is significant on two related fronts: First, it formally incorporates the fundamental aspects of Microsoft’s Azure cloud into an open community that has so far worked in and around Linux as the primary operating system, and secondly it adds another option to the OCP stack so that enterprises that choose to deploy it will not necessarily do so on the same hardware or software infrastructure.
The question on most people’s minds, however, is whether this is a sincere effort on Microsoft’s part, or is it simply hoping to blunt the OCP effort to give Azure a better chance to succeed on its own. Some IT analysts, like Gartner’s Jeffrey Hewitt, are less than impressed, suggesting that the move is mostly exploratory at this point and gives Microsoft an aura of openness while at the same time offering the possibility of more Windows-based hyperscale infrastructure once deployments get going in earnest. It will be a tricky dance, however, considering Microsoft has many lucrative bundling deals with the very server manufacturers that OCP hopes to supplant.
For the enterprise, however, it provides a chance to build private versions of the Azure cloud that can then support many popular business applications, such as Office 365 and SharePoint, which have ties to the legacy infrastructure powering most enterprise activity today. Along with server hardware and software, the package includes various CAD models and schematics, plus the source code for chassis functions such as server management, power/cooling designs and diagnostics. Anyone who has been watching Microsoft over the years knows that source code is not something the company gives up willingly unless a high likelihood of substantial returns exists on the back end.
But what exactly is the Azure design, and is it suitable for the kinds of hyperscale architectures the enterprise is looking for? A deeper dive into the Microsoft hardware shows a 52U rack rather than the standard 42U capable of holding four 24-blade enclosures in which each blade fits on half a slot. That means 96 servers can slide into a single rack, with Microsoft said to be deploying 10 dual-core Xeon E5-2400 processors per blade. The key difference with existing OCP designs is in the rack width, with OCP going with a 21-inch form factor that allows it to go three-wide with its blades, but on a 2U slot. As well, OCP reserves one RU for JBOD storage while Microsoft places storage within the blade. Both designs are highly modular and are geared toward scale-out infrastructure using only a fraction of the hardware that a typical deployment would require.
This is crucial because if there is one thing the enterprise hates more than not having enough resources to do the job, it’s not having enough money to pay its bills. The hyperscale movement is born of the need to ramp up resources for Big Data and Web-facing operations without blowing either the capex or opex budgets.
By joining OCP, Microsoft may make headlines today, but the real test is whether its architecture will fulfill the needs of tomorrow.