ARM aims at network infrastructure with many-core SoCs and open software stack

ARM details its vision for next-generation network infrastructure based on ARM chips and open-source software

Chip designer ARM has detailed its vision to drive next-generation network infrastructure, revolving around ARM-based processors and a common software stack to deliver greater intelligence distributed throughout the network to cope with the increasing data and traffic demands.

In a briefing ahead of Mobile World Congress (MWC) next week, ARM announced its plans for the "intelligent flexible cloud", and how it believes telecommunications providers can keep on top of future challenges by embedding in-network and edge-of-network intelligence, rather than trying to handle all processing in data centres or the end device.

The firm also revealed how licensee partners are delivering a new generation of high-performance ARM-based system on a chip (SoC) processors and equipment that are being designed to meet these challenges.

"Infrastructure today is still based on an architecture that is decades old. There's been bolt-ons and technology upgrades and tweaking in order to meet the need of our rapidly growing mobile world and broadband, but some critical factors have stayed the same," said Charlene Marini, vice president of marketing for Embedded at ARM.

"One is that there is still a reliance on single-purpose equipment, and the other is that application intelligence is still constrained to use the capabilities of an end device at one end of the network or a data centre at the centre of the network, and this has limited the amount of innovation that can take place on the network side of the infrastructure."

This is now being exacerbated by trends such as cloud computing and the Internet of Things, which are causing not just an explosive growth in data volumes, but also in the diversity and density of data. There is also the growing issue of the amount of energy being consumed by data centres and network equipment, Marini said.

There is also the question of latency. With upcoming 5G mobile networks specifying latencies below 1ms, sending requests all the way up to the data centre and back means that those requirements will be difficult to meet.

ARM believes that these issues can be addressed by building in intelligence to every tier of the network infrastructure, including the access layer and network edge, so that applications run wherever the data is.

The limitations on space and the availability of power where much of this equipment is located plays well with the size and energy efficiency of ARM chips, and hence the firm sees a big opportunity here.

ARM is bringing to the table a framework built around SoCs that can scale to more than 64 cores, with the ability to integrate different I/O and memory interfaces, plus custom logic keyed for specific applications.

"The ability to integrate accelerators and other types of intellectual property (IP) that can efficiently process workloads is going to be critical in this intelligent flexible cloud framework," Marini said.

Two partners already working with ARM under this framework are Freescale and Cavium.

Freescale's QorIQ LS2 family sports up to eight 2GHz Cortex-A57 64-bit cores, coupled with a network acceleration subsystem described as a "GPU for networking" by the firm, and eight 10Gbps Ethernet interfaces. It is aimed at applications involving software defined networking (SDN) and network function virtualisation (NFV).

Cavium's ThunderX line boasts up to 48 64-bit ARM cores running at up to 2.5GHz, and is aimed at a wide variety of data centre applications. Different chips in the family have different accelerators for storage, networking, compute or security applications.

The family also boasts a number of on-chip Ethernet ports that have the ability to be configured as a low-latency fabric for clustering nodes together.

Underpinning the hardware is a common software stack that will enable developers to create applications that are portable across the network.

This comprises largely open source software and standards, such as the OpenStack framework for orchestration and management, the Open Data Plane APIs, Open vSwitch, KVM and Linux.

"Open Data Plane is the key enabler. It's cross-platform and provides a common target for developers to access both hardware and software acceleration capabilities that are critical to meeting real network requirements," Marini said.