Emerging Processor Platforms Eager for SDN

Software defined networking (SDN) and network functions virtualization (NFV) are poised to remake enterprise data environments from the ground up, so it is no wonder that network processors are  quickly shoring up support for both formats.

Most of the management and configuration magic of abstract networking will take place in software, of course, but at some point the virtual world has to draw from the physical. It makes more sense to build atop an SoC optimized for SDN/NFV rather than to force-fit such functionality onto legacy silicon.

Two key advances recently came out of Broadcom and Freescale Semiconductor, which appear likely to become major rivals for the abstract networking market. Broadcom’s most recent contribution is the StrataXGS Trident Ethernet portfolio, led by the Trident II+ SoC, designed to layer virtual switch topologies atop 10 GbE deployments. The device’s switch capacity was recently upped to 1.28 Tbps, coupled with improved performance for network overlays. Freescale, meanwhile, is working with Advantech on white box solutions that enable low-cost, programmable network environments using open community platforms like OpenFlow.

Meanwhile, Intel is looking forward to the day when abstract network architectures are combined with modular infrastructure to provide highly dense but broadly scalable data environments. The company recently released the Xeon D chip, built on a 14 nm process that would provide a unified processing architecture across microservers, network devices and storage appliances. Initial designs feature quad- and eight-core devices with integrated dual 10 GbE ports, AES-NI encryption and the company’s Platform Storage Extensions data acceleration system.

The problem with increasingly dense hardware configurations, however, is that heat loads tend to ramp up to dangerous levels fairly quickly. To that end, some designers are working with optical networking modules that can sit directly on the motherboard to improve both cooling and power consumption. The Consortium of On-Board Optics (COBO) held its initial meeting in Santa Clara, Ca, earlier this month, led primarily by Microsoft in an effort to streamline hyperscale infrastructure. The group also includes Cisco, Juniper, Mellanox and Arista. COBO will work primarily on standard interfaces, thermal requirements and other criteria for the creation of interoperable modules for switch and adapter configurations.

This is all part of a concerted effort on the part of hyperscale cloud providers to propel network hardware to their required levels, says EE Times’ Rick Merritt. It started several years ago with the drive to 25 and 50 Gbps and is now gravitating toward greater flexibility and more hierarchical approaches to SDN through advanced labeling schemes and sophisticated packet handling. Chip vendors are already writing code to these new “Flexible Ethernet” standards, with initial trial deployments expected in the next few months.

To be sure, specialty silicon is not required to implement SDN, but it can certainly help to optimize it. Refitting network infrastructure is no easy task, but as legacy infrastructure makes the transition to advanced virtual data architectures SDN capability is expected to work its way into the data center via the normal hardware refresh cycle.

If the enterprise is interested in deriving the maximum benefit from software defined architectures, it will need to start with a strong foundation in silicon that can handle the dynamic workflows and rapid configuration requirements the network abstraction engenders. At the same time, SDN-optimized processors will do wonders to lower costs, ease management burdens and foster increasingly advanced applications and services.

Photo courtesy of Shutterstock.

Latest Articles

Follow Us On Social Media

Explore More