The new network processor will replace the router and switch

Nick McKeown, a professor of engineering at Stanford University, expects a new network processor to replace the ASIC currently used in routers and switches in the next decade; he said he has gone deep into the future of the Communication Processor: "and if you try to look at it with your eyes, it's like a network-based RISC processor."

McKeown assists in promoting a software-defined network based on OpenFlow communication protocols. Its goal is to generate a series of new software applications to manage various simplified switches and routers. If the task is successful, it will make it easier to operate a large data center and enterprise network at a lower cost, it is expected to subvert the current business model of network devices that uses complex ASIC and exclusive program code.

McKeown's new commercial chip family is expected to replace existing ASIC solutions for major manufacturers including Alcatel-Lucent, Cisco, Ericsson, And Juniper. He said, the first new chips of this type will be available in the next 2 ~ Three years ago.

Through cooperation with TI and other vendors, McKeown prototype the new product components in the form of a thesis. It basically contains various headers (headers) that are constantly added to each data packet for translation) the parsing engine (parsing engine), and then push the package into a execution unit (execution unit) pipeline that matches the graphics in the header, and make its action (action ).

"This is a mandatory match-action, match-Action feed-forward pipeline (feed-forward pipe);" McKeown says his research papers are under pre-publication Review.

According to McKeown's paper, as long as the chip area and power consumption increase by 15%, the new chip can achieve the same performance level as ASIC that currently only supports specific communication protocols, processes any communication protocol today or in the future. He expects that those large routers and switches will replace their ASIC with such chips within ten years and transform them into software companies.

"We can see the results within ten years, and those vendors will provide control plane software and applications at the top ." He pointed out that two or three companies are already researching and developing such chips, including new startups, xPliant, and established companies such as TI or maybe Cavium and Mellanox.

"Commercial chips will be one of the main drivers of the OpenFlow Program," McKeown said: "existing chip vendors, including Broadcom and Marvell switch chips, are already preparing to support OpenFlow-they should have done this and they have been involved from the very beginning."

With the birth of a new generation of software and hardware, technological evolution will promote the industrial revolution. Currently 1. the OpenFlow of Version x represents a compromise, McKeown pointed out: "Ideally we will let it begin with a universal matching-Action workflow, but it also needs to map it to existing chips-the next generation of technology will be less dependent on specific protocols."

Last year, the Open Networking Foundation (ONF) responsible for supervising the OpenFlow protocol was busy inviting ASIC suppliers to set up a Forwarding processing actions Working Group; the objective of the team is to narrow the gap between the goals that OpenFlow hopes to achieve and the existing and planned ASIC goals.

Now ONF is working on the establishment of the chip advisory board. McKeown said: "We will learn from them the possibility of chip technology, from there, we can extend the possibility of the next generation of OpenFlow."

OpenFlow has begun to use content-addressable memory (content-addressable memories) as an intermediate medium for routers and switches ASIC. However, the functions of this method are limited; therefore, the standard recently began to use a matching multiple tables technology. "Not only does the OpenFlow version support a single protocol take some time to come," McKeown said ."

In the software sector, Software Defined Network (SDN) program code is also being developed. New startups include Big Switch Networks and Nicira, which have been acquired by VMWare, they already have their OpenFlow controller software. Earlier this month, 18 large communication equipment vendors and software vendors announced cooperation to promote an Open Daylight program designed to provide SDN controllers and related application interfaces (APIS).

Industry observers expect major manufacturers to compete to make their program code part of the Open Daylight program. Once reliable products are available on the market, large businesses such as IBM will be willing to spend money to adopt them, to provide integrated services-but I'm afraid it will take two to three years.

McKeown uses a standard API set Posix that later became Linux as a metaphor; he pointed out that, it took a decade for the operating system of different versions to settle down to a stage in which Posix application interfaces can be written, the same situation may also occur on the SDN application interface.