Why 2016 may finally be the year of the ARM server

Technology editor Dan Robinson ponders whether ARM chips can break into the data centre market

ARM-based server systems have had a rocky road to adoption. It seems like many years since the ARM architecture was first mooted as a potential server platform, and since then we have seen numerous announcements from ARM and its partners about initiatives and developer programmes, but commercial products have proved almost as rare as hens' teeth.

ARM has been discussing plans for servers based on its architecture since at least 2010, when only 32-bit ARM processors existed, while data centres and server rooms were full of systems based on 64-bit x86 chips.

This didn't put off a startup called Calxeda (originally known as Smooth-Stone) announcing its intention to exploit the low energy characteristics of ARM chips and bring a server platform to market based on the technology. Sadly, the company ceased operations in 2013 after apparently running out of funding.

The following year, ARM announced the ARMv8 architecture that paved the way for 64-bit core designs in 2012, again talking up the potential of these for server applications. ARM does not manufacture chips but licenses its technology to firms such as TI and Samsung, so it took a year or two for this to filter through into production silicon.

Some ARM-based servers have seen the light of day, most notably HPE's Moonshot server platform which can be configured with ARM-based modules, but this targets very specific workloads and applications. Lenovo is likewise said to be looking at a similar strategy, and is reportedly developing its own ARM servers optimised for specific kinds of workload.

Now, AMD is stepping up to the mark and making its ARM-based Opteron A1100 server chips commercially available with a view to establishing itself as a leader in this area and to get ARM-based server chips adopted more broadly than in just a few niche applications.

AMD's involvement in the ecosystem brings a certain amount of credibility to the notion of ARM-based servers. The chipmaker is Intel's biggest rival in the processor market, and has already enjoyed a measure of success in servers with its x86 Opteron chips, a fact that means it already has relationships with server vendors.

However, even AMD seems to have had its work cut out in actually getting ARM servers to market. The Opteron A1100 was first revealed to the world in 2014, when it was made available as part of a development platform for partners, meaning the technology is already a couple of years old.

Many problems have stemmed from the fact that servers are a whole new use case for ARM chips. With x86 systems, which have been around for a long time, there are standard ways of initialising the hardware and booting the operating system, for example. Recognising this, ARM developed the Server Base System Architecture specifications to define a standard system with which hardware and software partners can work.

Then there is the unavoidable fact that x86 servers totally dominate the data centre market, and hence there is a vast library of operating systems, middleware and applications for customers to choose from with x86 systems.

To stand against this, many in the open source community are adding support for ARM-based systems, in particular the Ubuntu, Red Hat and Suse Linux distributions, and Linux is the dominant operating system used in the service provider market.

Meanwhile, the strength of ARM chips is in power efficiency, and many data centre operators have expressed concerns in recent years about the growing energy budget of their infrastructure and the associated costs of cooling all those hot servers.

This means that ARM-based systems could get a look-in for some applications, and AMD is targeting networking, software-defined storage and web server roles initially. In other words, the Opteron A1100 is going to start off serving specific workloads, like many other ARM chips.

However, it is still early days and, while few industry observers expect ARM chips to displace their x86 rivals from data centre dominance, we may eventually look back on 2016 as the year that ARM-based servers finally gained a foothold.