Intel Corp. threw its weight behind software-defined networking and network functions virtualization with a pair...
of reference designs for white-box switches and servers that could help mainstream white-box hardware.
The reference designs will help original design manufacturers (ODMs) build hardware that supports both physical and virtual data forwarding elements for software-defined networks. Intel also revealed its Data Plane Development Kit (DPDK), a new programmable forwarding module and application program interface (API) for Open Virtual Switch (Open vSwitch) that optimizes virtual switching. Intel DPDK will also be the software foundation for Intel's server reference design.
"If there is going to be anything like an OpenFlow value chain, it has to start with merchant silicon. It's got to be out there with a solid and well-defined roadmap so that OEMs and ODMs can adopt it and [so there's] enough demand for them to make the product," said Brad Casemore, research director at IDC. "Intel is trying to make everyone comfortable that if you adopt their technology you are on a firm foundation. I'm sure we'll see Broadcom make some announcements soon."
The Intel DPDK and reference designs could also help unify the physical and virtual switching that takes place in many networks today. "This provides an interesting shift in terms of trying to bridge that gap [between switches and servers] and trying to provide value by getting equivalent capabilities in either direction," said Eric Hanselman, research director at 451 Research.
Intel's focus: Open infrastructure
Intel believes that SDN and network functions virtualization, or NFV nodes need to be based on open platforms, according to Rose Schooler, vice president of the Intel Architecture Group and general manager for the Communications and Storage Infrastructure Group. "We don't believe you should overlay it on existing proprietary implementations," she said. "We believe in openness of northbound and southbound APIs and interfaces. And we believe you need a reference design to spur that new way of thinking."
Intel has an early set of partners for its reference designs that are well-established in the SDN industry, including vendor partners Big Switch Networks, HP, NEC and VMware; and carrier partners, such as NTT and Chungwha Telecom. AT&T has commenced a proof-of-concept project with the reference designs, Schooler said. ODM Quanta also has its first production deployment of a switch based on the reference designs, she said.
Intel Open Network Platform Switch
Intel's switch reference design, Open Network Platform (ONP) Switch is based on a pair of silicon platforms, Intel's Ethernet Switch 6700 series and its Communications Chipset 89xx series. It also includes a new network switch software stack based on Wind River Linux, and supports OpenFlow and Open vSwitch.
The reference design will enable manufacturers to build top-of-rack switches that support OpenFlow 1.0 and tunneling protocols like Virtual Extensible LAN, or VXLAN, and Network Virtualization using Generic Routing Encapsulation, or NVGRE.
Intel DPDK for Open vSwitch and the ONP Server
Intel's DPDK is a rebuilt forwarding module for Open vSwitch that optimizes and accelerates small-packet throughput, performance and packet processing. Schooler said Intel will make its DPDK available to the open source community for consideration for inclusion in the Open vSwitch project, which VMware controls.
Allwyn Sequeira, vice president of networking and security at VMware, said his company is working with Intel to roll some of DPDK's enhancements into Open vSwitch.
DPDK is the foundation of the ONP Server, Intel's reference design for SDN-friendly servers. Based on the Xeon processor, the Intel 82599 Ethernet Controller and the 89xx series chipset, ONP servers will have a software stack similar to Intel's ONP Switch. The server will use DPDK to provide efficient virtual switching. It will also serve as a platform for network functions virtualization by providing an x86 platform that is optimized for computationally intensive functions, such as deep packet inspection and fast packet acceleration.
Having an optimized vSwitch and optimized server hardware for virtual networking will help data centers avoid burning up all their CPU cycles on computationally intensive network processing. "If you have lots of servers with Open vSwitch, you have all this acceleration to give you high performance," said Hanselman of 451 Research.
Intel's server and switch reference designs also include a common API that will simplify how SDN controllers manage both virtual and physical switches, an issue that has come into focus with other vendors. Arista Networks recently submitted code to OpenStack to unify how Quantum orchestrates both physical and virtual switches.
"We look at enterprise and cloud data centers as the natural proof point [for SDN,]" Intel's Schooler said. But data centers are just the tip of the iceberg. "We've looked at all those use cases for service provider SDN and network function virtualization," she said. Intel sees its reference designs as helping service providers with their access, edge and core networks, including base stations, border network gateways, media gateways and evolved packet cores.
Verizon Communications is preparing to trial hardware based on Intel's reference designs, according to Prodip Sen, Verizon's director of network architecture. "Why are we doing this? We need to lower our costs," he said. "Traffic is growing on our networks exponentially, and we cannot afford to build networks as we have in the past. If we do, we'll go out of business."
Verizon is also preparing an SDN-based, data center-to-data-center trial with Intel to test cloud bursting between an Intel lab in Portland, Ore., and a Verizon lab in Waltham, Mass.
Let us know what you think about the story; email Shamus McGillicuddy, News Director.