ARM chips have long been favoured by mobile device manufacturers, as their reduced instruction set computing (RISC) architecture means they consume power, generate less heat and – in turn – cost less to run.
Unsurprisingly, these characteristics haven’t gone unnoticed by the cloud and datacentre industry, but it’s taken a while for the technology to really get its foot in the door.
ARMv8 was the first ARM architecture to include a 64-bit instruction set and made its debut way back in 2011. Its energy efficiency and performance was deemed a good fit for datacentre operators looking for a cost-scalable way to deal with the rising demand for cloud services.
Progress in this area has been slow, though, despite the fact the technology’s power consumption – which has fuelled its success in the mobile market – is far lower than the equivalent generation of Intel offerings.
That being said, the energy efficiency of ARM 64-bit systems can only really come to the fore if the applications that run on them can make correct use of the architecture.
“If they don’t then there may be no difference to x86 in terms of energy usage, and to achieve a given number of computations in the same time frame may actually require more ARM cores than x86,” says Ark Data Centres chief technology officer Jason Liggins.
“This along with the cost of porting applications from x86 to ARM may mean that there is no compelling reason to port from x86 to ARM.”
Mobile technologies in the datacentre
David Leyland, UK and Ireland head of Dimension Data’s next-generation datacentre business unit, explains the technology also gives server makers the ability to add new features and functions to the chip.
“One of the key beauties of ARM for datacentre environments is that the chip design is licensed. Hardware manufacturers can add specific functions around the core processor,” he says.
“In the same way that, for example, Apple adds accelerators for iPad functions, server manufacturers can add peripheral circuits to optimise network transfers and so on, and doing so can speed up collaborative computation.”
However, while this sounds good on paper, it can be difficult to do in practice because the design philosophy for servers and mobile devices are so different.
For example, the former favours embedded designs that fuse the software to underlying hardware, which works incredibly well in smartphones and tablets, but less so in the datacentre world where the hardware and software inside may come from two different suppliers.
Datacentre mix and match
Clive Longbottom, service director at analyst firm Quocirca, says mixing 64-bit and 32-bit workloads in the Intel world is relatively difficult, but is easier to do when using ARM AArch64 chips.
This is important, he adds, because there is a lot of ARM code out there, still in use, which will have been written for the older ARM 32-bit architectures.
“Therefore, I would expect that the benefit of this is in being able to port old code over easily while creating new code. This could lead to an easier migration as the old stuff gets re-written to being 64-bit,” says Longbottom.
Also, while Intel x86 chips are a good fit for workloads where high-powered systems are required, there are instances where using a less powerful system to carry out the job would suffice.
“There are many other workloads that could be run on systems that are less powerful and ARM provides an environment where such workloads can be run on low-power, low-heat, low-cooling servers,” Longbottom says. “Intel is trying to counter this with its Atom-based servers, but is a bit behind what ARM is now offering."
The introduction of ARM in the datacentre could herald growth in the types of silicon being used in the cloud, Longbottom claims. “OpenStack is moving to being a multi-silicon platform, through the use of x86, Atom, ARM, Power, Sparc and so on, so being tuned to deal with the differing needs of different workloads becomes more important," he says.
“However, intelligent software will be required to ensure that the right workload is in the right place at the right time.”
OpenStack’s work in this area also has the potential to make life a little bit easier for those already using ARM in the datacentre.
“For those who have gone for ARM in their own datacentre, they want to have cloud-based systems where they can develop, test and run workloads against the same ARM environment. Now that OpenStack runs on ARM, it opens up options for providers in different ways,” Longbottom says.
Tony Lock, distinguished analyst at market watcher Freeform Dynamics, says for operators that have to deliver services to widely distributed users and service providers, 64-bit chips can be a big help.
“This concerns the ability to have a common chip platform of various scales positioned in the appropriate locations – datacentre class chips in the centre and very low-power solutions close to the users at the edge of the distribution network, for example” he says.
Datacentre drawbacks of 64-bit ARM
While the benefits of 64-bit ARM in the datacentre might seem alluring from a power and efficiency point of view, it does have some downsides, according to Dariush Marsh-Mossadeghi, director of technology strategy and architecture at co-location firm DataCentred.
“There is currently a lack of maturity in certain elements of the software stack that makes production systems challenging today," he says. "While the processor support is all there and works well, the challenges are in the peripheral aspects of the hardware, bringing – in turn – obstacles in the areas of basic input/output systems, boot-loaders and low-level kernel features.”
Read more about ARM in the datacentre
This is a viewpoint product strategy director for Virtus Data Centres product strategy director Matthew Larbey shares. “Many of the operating and application stacks will need to be redeveloped to be able to really tap into the architecture benefits," he says.
“It will be a chicken-and-egg situation where there needs to be a groundswell on both sides – hardware and software – to make the new server alternatives a success in the marketplace and truly offer a cost and performance alternative.”
Aaron Sullivan, senior director and distinguished engineer at managed cloud hosting firm Rackspace, says 64-bit systems may also require different approaches to fault management, storage management, monitoring and field service, which – again – could be another barrier to adoption.
“If your operations teams or system administrators are very tied to their existing redundant array of independent disks controllers, network adapters, management tools, and datacentre service models, they may need to change to incorporate ARM. The features are generally there, but the processor configuring and managing those features may be different,” he says.
Future developments
Looking ahead, Matt Lovell, chief technology officer at cloud hosting firm Pulsant, says wider adoption of 64-bit ARM in the datacentre could pave the way for greater integration between user devices and connected services.
“This enables a richer experience for the user – dependent on connection type, stability and strength – to be assessed and delivered,” he says.
“It also enables offloading of processor-intensive requests and applications to be seamlessly developed, which extends the life of user devices and interface investments, as well as removing constraints to provide a consistent experience regardless of device specification.”
Marsh-Mossadeghi, meanwhile, predicts that as collaboration between the hardware suppliers and the open source community continues to grow, the maturity of offerings coming from service providers such as DataCentred will also increase.
“This will manifest itself as providing access to a broader choice of operating systems and kernels on ARM-based cloud platforms,” he says.
