Broadcom Unveils Faster Data Center Switch Platform

With IT organizations starting to deploy as many as 30 virtual machines per physical server demand for network bandwidth within the data center is increasing rapidly. To address that issue Broadcom unveiled today an upgrade to its System-on-Chip (SoC) platform that can deliver 1.2 terabits per second of performance in a top-of-rack data center switch.

Dubbed the Trident-II+ Series within the Broadcom StrataXGS Trident Ethernet switch portfolio, the new SoC will enable manufacturers to start delivering as early as this year a new generation of switches that are not only twice as fast but also consumes 30 percent less power.

Rochan Sankar, director of product management and marketing for Broadcom’s Infrastructure and Networking Group, said the Trident-II+ Series is designed to be upgradable within the existing data center switch architecture employed by most vendors. As a result, switch vendors do not have to design a new chassis to bring the Trident-II+ Series to market.

Sankar said that over the next several years the bulk of data centers will be making the transition from 1G to 10G Ethernet. As they make that shift, the cost advantages of employing data center switches based on merchant silicon rather than proprietary ASICs will increasingly become apparent, he said.

“This platform is really targeting the long tail of the enterprise market that is still moving off of 1G Ethernet,” Sankar said. “We think that transition will be playing out over the next three to four years.”

Like the previous generation of Trident series switch platforms, the Trident-II+ is optimized to process VXLAN traffic generated by VMware virtual machines. Rather than relying on servers that need to allocate as much processing horsepower as possible to applications, the Trident series enables the processing of VXLAN traffic to be offloaded to the switch itself. The end result is more balanced data center environment, Sankar said.

IT organizations don’t tend to upgrade switches as often as servers. But there is enough critical mass starting to build in terms of the number of virtual machines deployed on each physical server to start forcing the issue. In fact, the rate at which VM deployments are growing suggest that IT organizations might be upgrading switches more frequently in the years ahead than they have done historically.

As that trend continues to play out, Broadcom is betting that the cost advantages of switches based on merchant silicon manufactured in volume are likely to have an impact on switch economics as x86 processors did on the servers those switches are inevitably connected with.