Microsoft’s new server design aims to make cloud infrastructure cheaper
Seattle — Microsoft is offering up a new server design that it hopes will influence the rest of the data-centre industry — a move that keeps pressure on hardware vendors such as Hewlett Packard Enterprise to lower the price of cloud infrastructure.
An earlier design by Microsoft now accounts for 90% of the company’s data centre server purchases, underscoring just how far it has moved in favour of its own specifications, often manufactured by commodity server vendors.
The new design will be submitted to the open source Open Compute Project for others to work on and use, while Microsoft itself plans to deploy it in data centres by the middle of next year, said Kushagra Vaid, a Microsoft GM for the group that engineers the company’s cloud hardware. It is intended to enable companies to deploy a wide variety of applications and can use different international power standards without having to alter the hardware, he said.
The goal is to ease cloud deployments, as well as collaboration on these kinds of open source projects. Microsoft is unveiling the design at an earlier stage than previous Open Compute projects to enable many parties to co-operate on it. The public nature opens the process to commodity server makers and brands such as Hewlett Packard and Dell, intensifying margin and market share pressure on those companies as many cloud operators forego brand-name hardware.
Since 2014, 90% of the servers Microsoft has purchased for use in its datacentres have been based on Open Compute design. Microsoft often buys these from what are called original design manufacturers (ODM), makers of generic, commodity hardware such as Quanta and Wistron, and the company worked with an unnamed ODM for the new design.
Microsoft also purchases from brand-name sellers, called original equipment manufacturers (OEMs), but those vendors must use Microsoft’s design and meet its price requirements. Microsoft likes the arrangement because it gives it a greater choice of suppliers and lets the company customise what it needs while boosting competition among vendors.
Other operators of massive data centres such as Amazon, Google and Facebook also opt for the often cheaper, unbranded servers, in some cases to the exclusion of the name brands. This dramatic shift in the server market, helped along by the open sourcing of cloud server designs, has given the US hardware behemoths of yesteryear a choice of two bad options, said Anand Srinivasan, an analyst at Bloomberg Intelligence.
"You participate in the business and its large revenue source but at very low margins, or if you don’t participate, you miss out on the opportunity, but your margins stay solid," said Srinivasan, who previously authored a report about the conundrum titled Darned if you do, darned if you don’t."
For Microsoft’s part, Vaid says the decision to publish the new design earlier than usual — it’s still about eight months away from being finished — helps all vendors and any customers who want to use this design because they have plenty of time to modify it and contribute their own building blocks. Vaid will announce the work on Tuesday at a digital infrastracture conference called Zettastructure in London.
Dubbed Project Olympus, the design is for a standard 19-inch server rack and has two significant changes from previous work: the first is a universal motherboard that can accommodate a wide variety of applications and workloads from artificial intelligence to big data to whatever the next big thing is three years from now. It is designed to be modular, so users can tweak it and add additional processors, storage and components as needed, Vaid said in an interview.
The second change is a universal power connection unit that enables a rack of servers to be plugged in to any of the different types of electrical connection and standards in a variety of countries. Currently Microsoft and others have to rejig the servers’ power connection for different countries. This change will save about four weeks of work when Microsoft and or others are rushing to deploy systems globally, Vaid said.
"This will help accelerate the growth of cloud computing," he said. "Because of the productivity gains and the time to market agility that will come about."