ARM Server Update
An explosion in data
The incredible rise in smart devices shows no signs of abating. Over one billion smartphones were shipped in 2013, hundreds of millions of tablets were sold, and smart devices – ranging from anything between televisions and wearable technology – are set to become truly mainstream. Powering the vast majority of these devices, and many more, is CPU and graphics technology from British company ARM.
A proliferation of mobile devices and increases in networking and broadband speed means that more data is produced than ever before, growing year on year. It has been estimated that, in 2017, up to 10 exabytes of data will be generated every month from phones alone. Putting this in context, if each byte was to represent a grain of sand, that’s more than all the grains on the entire planet!
Massive amounts of data will also be generated by other ARM-powered smart devices, from set-top boxes through to wearable devices, not to mention bandwidth required for online videos or by social-media websites. Any smart device will be part of the ‘Internet of Things’, and if it’s smart, it produces data. Putting it simply, efficient data processing, within datacentres, is going to become incredibly important, as the ‘Big Data’ industry is currently finding out.
Power and cooling are key concerns
Computers running inside datacentres have historically used x86-based technology for mainstream processing. This has worked well enough thus far, with increasing compute density and ever-faster chips, but escalating power and cooling costs have hamstrung the potential within datacentres. These two factors play a vital role in determining the design and construction of modern datacentres, so much so that any savings here are arguably more important than increases in processor performance alone.
Rather than employ general-purpose servers that are adequate for all types of data processing, companies are increasingly looking towards application-specific servers that target the exact need. Workloads such as cold storage, web serving, memcaching, and Big Data don’t require huge monolithic x86 servers and their preponderance on processing complex instructions and reliance on memory bandwidth. Rather, low-power, optimised servers can do the job just as well and provide significant total cost of ownership savings. This is where the simpler, leaner ARM reduced instruction-set architecture (Risc) architecture comes in once more.
Expertise in low-power processing, born from years of experience gained in designing architectures that power mobile devices, gives ARM a key edge with respect to energy efficiency. The ARMv8-A architecture is fully 64-bit compliant and, when used in a multi-core, multi-chip configuration, is an ideal candidate to run in the kinds of server farms and datacentres that cloud-based companies such as Facebook and Google are actively seeking.
The ARM benefit
The ARM advantage in low-power processing is very real, with power and cooling requirements – two key ingredients in the total cost of ownership – dropping by as much as 80 per cent when compared to rival solutions from Intel that use x86 processing. It makes implicit sense to use the right hardware, for the right workload, all of the time. ARM processors are specifically designed to scale across many cores and thus run highly efficiently in what are known as hyperscale datacentres that are increasingly in popularity.
Compelling hardware is only one part of the puzzle, however. The continued success of x86-based datacentres is inextricably linked to the software that’s run on the servers. ARM’s Risc uses a different type of software library, meaning that popular applications and workloads need to be rewritten in order to take advantage of the hardware. This is a potentially costly and time-consuming exercise for software vendors, underscored by the dearth of native 64-bit ARM-compatible applications, so support from industry leaders such as Microsoft is key.
There has been considerable progress on the software front in the last year or so. Important software providers such as Red Hat, SUSE, and Ubuntu, run on millions of servers worldwide, have baked-in support for 64-bit ARM processors, thus enabling small- to medium-sized businesses the opportunity to evaluate a non-Intel platform that provides better energy efficiency and a provably lower cost of ownership.
Yet it can be successfully argued that datacentre owners oftentimes own the software stack and optimise workloads for hardware. This means that, should they decide to switch over to ARM-based infrastructure, porting over software isn’t a significant problem.
ARM understands that long-term incursion into the datacentre space isn’t going to occur overnight. Appreciating that the datacentre opportunity needs to be thought of holistically, meaning more than just hardware excellence, it has created a Server Based System Architecture (SBSA) specification for 64-bit computing and hopes that it will enable simpler deployment for datacentre owners who see obvious merit in the low-power processors.
ARM is also pushing for more standardisation for operating systems, such that rolling in new patches isn’t the headache it can be when a number of disparate OSes are run in one environment. Ease of deployment and management are key hurdles that ARM and its partners need to overcome when changing the status quo, as well.
Provable technology with major backing
Companies such as HP and AMD have seen the clear power and performance benefits of ARM-based datacentre technology. AMD has developed the Opteron A1100 processor, based on ARM Cortex-A57 cores, for this purpose. This AMD system-on-chip (SoC) is able to provide 2.5x the processing ability of an x86-based equivalent server CPU at the same power level. These kinds of numbers make every company who’s interested in data management sit up and take real notice.
And ARM is bullish that not only can its server architecture offer a compelling choice in the microserver market characterised by dense computing, it can also compete with Intel in the mainstream, high-revenue server market because of the scalability of the ARM v8 architecture and innovation from partners.
For example, industry giant Qualcomm has announced that it is to build a 24-core server CPU based on the latest ARM architecture. Such a proclamation of intent is irrefutable evidence of the appeal for ARM-based servers for mainstream as well as entry-level datacentre workloads.
PayPal, too, is readily adopting ARM-based technology. In May 2015, it announced that the ARM ecosystem provides significant speed-ups whilst reducing the total cost of ownership. For example, evaluating the cost of processing the complex algorithms associated with fraud detection, which is required in real time, moving away from an Intel-based ecosystem to an ARM one reduced acquisition costs by 9x, power consumption by 2x and increased rack density by a factor of 7x, according to the company. Such meaningful benefits are impossible to ignore for a business that wants to run efficiently.
Scalable, power-efficient processing is the way forward
And it’s not just in pure servers that ARM is a good choice, either. The growing needs of data management require more investment in networking and infrastructure – an area in which ARM-based designs are already commonplace. It is likely that networking and processing will be tied more closely in the future as data throughput becomes a key concern. For example, as more devices are connected to the Internet, wired and wirelessly, all this traffic needs to be processed somewhere, most likely a server farm, and there exist untapped opportunities for more low-power hardware integration. Datacentres will change to accommodate this new paradigm.
ARM’s collaborative approach also works better from an SoC point of view. Manufacturing partners are free to add in the supporting features of their choice, be this be dual 10Gig Ethernet, as is the case with the AMD Opteron A1100, or fully-custom, high-performance cores that will address a larger segment of the datacentre market, as proposed by Broadcom.
The ARM argument for the datacentre is disarmingly simple. The low-power architecture, lower total cost of ownership, seamless scalability and excellent per-watt performance ensures it’s a great fit for a large number of modern datacentre applications. There are billions of ARM-based processors in the wild; they’ll need billions more to serve as the backbone of the Internet. If managed and implemented correctly, with help from its wide range of industry partners, there is every reason for ARM technology to gain a larger footprint in the lucrative datacentre industry. Why pay for more when you don’t have to?