Microsoft is unlikely to use ARM-compatible processors in a meaningful way in its data centers – unless there is a huge change in the software ecosystem around the non-x86 chips, a top Redmond bod told The Reg.
Though Microsoft is closely watching developments in the ARM world, it's unlikely that the Windows giant will be one of the early large-scale adopters of ARM-powered servers, the company's general manager of Windows Azure Mike Neil indicated.
"It's a new technology, but where is it going to be disruptive? A big challenge ARM has is what workloads are you going to run on it," Neil told us.
His comments follow news that Google and Facebook are investigating the chips for major production use, and the combustion of ARM server specialist Calxeda which bet too much too early on 32-bit ARM being ready for the data center.
As Microsoft shifts from being a rowdy collection of independent business divisions and technologies into a more centralized "devices and services" company, the firm is faced with a problem that other large-scale web operators are free of: customers expect Microsoft to stress-test the software it sells within its own infrastructure before selling it to the public.
This means that if Microsoft made a shift to ARM, it would have to make sense both from an infrastructure standpoint, and a product standpoint.
Amazon Web Services faces a similar problem: why go to the trouble of buying ARM-powered servers, and port the AWS cloud orchestration software, if there isn't a large base of applications for customers to run to take advantage of the fledgling hardware platform?
This is a problem that Google, Facebook and such web giants lack, as they are working just for themselves. They make their money from advertising served by their systems, and so can use ARM's customization opportunities to reduce costs and accelerate particular tasks as required. They don't have to worry about selling the tech to others.
Of course, unlike Amazon or Google or Facebook, Microsoft is rather well positioned to kickstart an ARM-on-servers software ecosystem by building Windows Server to run on the chips, but this brings its own challenges.
'ARM creates a discontinuity'
As Microsoft grows its empire, applications ranging from Xbox game backends to Bing queries to Hotmail mailbox processing or Office365 workloads all jostle for space in Redmond's large cloud, a pool of underlying common hardware.
The worry is that adding in ARM to a global pool of more than a million servers could create ricocheting complications as tasks juggled between x86 and ARM CPUs. The operating system must be able to schedule processes to run on the silicon best designed for the task at hand: various different ARM system-on-chip packages in a cluster could be customized to accelerate particular tasks, so the kernel would have to take that into consideration.
Deploying ARM tech also messes with the economies of scale Microsoft enjoys from being a mega-buyer of chips, and could force up the price it pays for its x86 silicon.
"There's enormous value to us to have fungibility, and if I specialize my hardware for a very specific task, if I put that hardware into the pool with [my resource scheduler] and another task wants to go there, it can't," Neil explained.
"So that's one of the reasons why you see all the large players having large generally homogeneous pools of resources. ARM creates a discontinuity there.
"It's like any industry," he added. "A million to do a compiler, $10m to do ports, $100m to do apps, probably $1bn to make an ecosystem that is self-sustaining. Industry has tried that a few times: we've had Itanium in the past, it's expensive."
Itanium was Intel's bet that it could introduce a new architecture and the performance gains would be good enough that the software industry would create new applications for it. This didn't work. Adoption was muted, and HP ended up being the last major IT company to have a big Itanium business, and is in the process of retiring it.
When you're a company the size of Microsoft, you don't want to make a chip architecture bet and get it wrong.
"[ARM] will go through that maturation cycle," Neil said. "It'll be interesting to see if they make it all the way through."
If ARM chips do come to the data center, then it's likely Microsoft will fiddle with and perhaps embrace them, but judging by Neil's comments Redmond is not going to be one of the major data center firms actively pushing the low-power chips into the modern data center – unlike Facebook or Google. ®