Intel aims at enterprise, carrier networks with faster silicon
Intel has beefed up its networking silicon in a bid to capture a bigger share of enterprise, carrier, and cloud-provider networks for its chip architecture.
On Wednesday, the company announced a networking platform code-named Highland Forest, combining an existing Xeon CPU and the new Coleto Creek chipset for new levels of performance in several specialized functions as well as throughput.
Intel says a common x86 architecture can help networking vendors build new, more efficient types of infrastructure for enterprises and service providers. Rather than developing and refreshing a wide range of specialized boxes using different chip architectures, vendors can start to move toward x86 and a single software platform throughout most of the network, slashing the time and cost required to come out with new gear, said Rose Schoolar, vice president and general manager of Intel’s communications and storage group. She spoke on a conference call Wednesday to announce the platform.
The company has been pitching x86 as a superior alternative to proprietary networking chips and equipment for years and has made some inroads, especially with control-plane chips in switches and routers, according to Linley Group analyst Bob Wheeler. NFV (network functions virtualization), designed to shift tasks in carrier networks from specialized devices to standard, virtualized server hardware, could help Intel to propagate x86 further into networking, Wheeler said.
Carriers want to match the flexibility of providers of so-called over-the-top Internet services by having similar virtualized data centers. But some network functions still lend themselves to specialized hardware, so there are limits to what Intel may accomplish there, Wheeler said. “It’s not as simple as it sounds,” he said.
Networking has lagged behind hardware standardization and virtualization in the world of computing. Intel itself estimates that less than 5 percent of network infrastructure is virtualized.
Intel’s networking chipsets offer a common set of functions across Atom and Xeon solutions for small to large systems. Those functions include IPsec and SSL security, Intel’s DPDK (Data Plane Development Kit) technology for faster packet processing, and Hyperscan deep-packet inspection powered by technology Intel acquired earlier this year with Sensory Networks.
Highland Forest sits at the top end of this lineup. Like Intel’s previous offering, a refresh of the Crystal Forest architecture, it’s based on the Xeon E5-2600 v2 processor. But a typical configuration of Highland Forest will consist of two of the 20-core processors and four Coleto Creek chipsets. It’s designed to take on specialized networking silicon with high performance: 225 million packets per second of throughput, Hyperscan deep-packet inspection at 140G bps, up to 110G bps IPsec processing and up to 200G bps of SSL encryption performance, according to Intel.
The new platform benefits from improved performance in the Coleto Creek chipset compared with its predecessor, Cave Creek. Among other things, the new chipset can deliver as much as 40,000 operations per second of RSA decryption, versus 5,500 operations per second for Cave Creek. It can also compress data at 24G bps (bits per second), up from 8Gbps for Cave Creek.
Intel’s biggest edge in achieving this performance is its ability to optimize software functions, such as Hyperscan, for the x86 architecture it knows so well, Wheeler said.