Experts see promise, threat in SDN, OpenFlow

SANTA CLARA, Calif. – A panel of experts gave mixed reviews on software-defined networking and the OpenFlow protocol at the annual Hot Interconnects conference here. Much of their conversation revolved around the potential of SDN to undermine current business models in communications.

SDN aims to shift the job of controlling networks to widely used programming environments on servers, eliminating a growing pile of complex protocols run on ASICs. OpenFlow is a specification for servers to control the operation of packet-forwarding tasks in routers and switches.

Panelists generally agreed the broad trend to SDN is significant, but many raised doubts the still nascent OpenFlow technology is the best implementation of it. They all raised fears the trend could puncture the profit margins in communications as it opens up a new businesses for developing server apps that control network functions run on simplified routers and switches.

“High [profit] margins are needed to support high innovation,” said Kireeti Kompella, chief technology officer at Juniper Networks. “Yes, we are stuck with Internet Protocol, but we have innovated around it so there’s room for competition,” he said.

Kompella characterized much of the thinking around SDN as a re-hash of prior work on network virtualization. “It’s not a simple decision between the flexibility of a virtual switch [in software] and purpose-built ASICs that do forwarding,” he said, referring to the ASIC-rich systems of companies such as Juniper and rival Cisco Systems.

Dave Meyer, a distinguished engineer at Cisco, took a middle-ground position, depite the fact his company to date has not directly embraced SDN or OpenFlow. “OpenFlow challenges most of our central dogma…[but] the [SDN] genie is out of the bottle, and it’s probably not going back anytime soon,” he said.

“I don’t think OpenFlow will kill switch ASICs, it will just re-aim them,” said Jeff Mogul, a fellow at HP Labs.

Mogul was bullish on SDN, but said OpenFlow is just one, relatively immature implementation of it. He noted more than 180 researchers from 90 groups and 17 countries attended a recent SDN technical forum in Helsinki.

Asked whether it’s a fad of phenom he said, “My sense is it’s a fad with legs, if it’s a fad at all…I am sure I could be doing SDN research for the rest of my career.”

SDN is initially aimed at large data centers such as Google trying to simplify their massive networks. However Mogul said HP has one carrier customer who is working with SDN, and Google is working with telcos in other countries to get them to adopt the OpenFlow technology it now runs internally.

Companies such as HP and IBM see less threat and more opportunity in the move to SDN. That’s because they have server, software and services businesses they aim to grow and relatively small switch and router businesses that could be disrupted.

“The key thing for IBM is apps-aware networks--let the app tell the net what it wants,” said Vijoy Pandey, chief technology officer for IBM’s network OS group. “There’s no virtualized provisioning of a network in the data center, and that’s becoming a real problem,” he said.

“I believe in SDN, but I am not sure [about] OpenFlow,” said Dimitri Stiliadis, a senior director of the IP division at Alcatel-Lucent.

He described OpenFlow as just one way—and not necessarily the best one--to implement one aspect of an SDN. He added that the move to SDN could spawn a new class of network integrators who combine SDN software developed by one set of companies with new simplified routers and switches made by other companies.