The enterprise is about to see a flood of ARM-based networking devices in the channel, but aside from the lower power consumption, how exactly will they change network environments for the better?
For starters, they should help to break the stranglehold that MIPS and x86 solutions have had on the network SoC field for so long. The real benefit, however, should come in the ARM’s ability to better handle the small-packet, bursty traffic that accompanies an increasingly mobile, cloud-savvy workforce.
Top network vendors like Cisco, Netgear and HP are all moving aggressively to deploy the ARM architecture throughout their portfolios. A top choice is AppliedMicro’s 64-bit Helix solution, which is based on the same design as the X-Gene that currently powers HP’s Moonshot server line. The Helix, however, is optimized for both networking and storage, as well as key embedded solutions like printers and industrial devices. Initial Helix designs offer four cores running at 1.2 GHz, although eight-core, 2.4 GHz solutions are currently under development.
Meanwhile, Freescale Semiconductor is out with the new QorIQ LS1043A quad-core chip, which the company says is designed to provide a high degree of intelligence on the network edge to boost the performance of virtual and software defined architectures. The device is based on the 64-bit ARM Cortex-A53, offering 1.5 GHz performance for as little as six watts. It offers on-chip support for network security and authentication, as well as virtual machine functionality and application visibility to gauge key performance requirements like latency and bandwidth support. In this way, the network plays a more active role in optimizing the user experience, rather than simply shuttling bits from point to point.
And AMD, which has made no secret of its support for ARMs in the server farm, is now turning its attention to the network. The company recently unveiled a specialty device aimed at its emerging Network Function Virtualization (NFV) platform, which ultimately is expected to unite ARM-based virtual networks with the installed base of x86 infrastructure. The Embedded R-Series SoC supports key aspects of the virtual network, such as traffic migration and packet management, with the idea that networks can then be constructed around software-based standards rather than proprietary hardware. The chips are currently in the hands of key developers like Vodafone and Aricent, with actual products expected next year.
While most initial ARM solutions will feature four or even eight-core designs, it shouldn’t be long for denser options to emerge. TSMC recently announced the development of a 32-core solution using a 16nm FinFET process that can then occupy the same substrate that houses a 28 nm Cortex A-57 device. The company calls the technique CoWoS (Chip-on-Wafer-on-Substrate) and allows for powerful networking solutions of extremely small size and low energy consumption. The system was developed in conjunction with Huawei’s HiSilicon division, although there is no word yet on when a commercial product will hit the channel.
It is clear, then, that even as network architectures continue to reach new heights on the virtual and software layers, there is still a lot happening on basic silicon as well. Things could get rather complicated when attempting to provision the right processors for select networks, but enterprises with a clear idea of what they want to accomplish should soon have an easier, and cheaper, means of getting there.
Photo courtesy of Shutterstock.