IBM became the second large IT vendor to introduce its own OpenFlow controller last week, announcing the availability of its Programmable Network Controller. Any doubts about IBM's intent to take on Cisco and Hewlett-Packard Co. in the networking market have been erased.
By submitting your personal information, you agree that TechTarget and its partners may contact you regarding relevant content, products and special offers.
Just a day after HP Networking announced its own OpenFlow controller, which will ship next year, IBM announced it is shipping its Programmable Network Controller (PNC) now.
IBM's OpenFlow controller software is based on IBM intellectual property and runs on an x86 server. IBM licenses the software in configurations to control one switch, 10 switches or 50 switches, and intends to offer a license to manage up to 100 switches soon. The controller features Web-based, RESTful northbound application programming interfaces (APIs) for the development of network applications on top of the controller. It also ships with topology discovery, fault detection and redundant configurability. IBM recommended that customers run a backup PNC in memory synchronization with a production PNC via two bonded Ethernet channels.
"With OpenFlow, we've eliminated Layer 2 design challenges and removed spanning tree. Any topologies can be supported with this capability. We also allow the commoditization of the network where customers can build logical topologies on top of their physical gear independently, so there can be changes to the vendor environment without affecting business logic," said Stewart Raphael, business line executive for IBM System Networking. "We see OpenFlow with PNC as a way that customers can create an open network environment where they can innovate and bring new applications online."
With OpenFlow controllers, networking vendors can develop application ecosystems
OpenFlow's governing body, the Open Networking Foundation (ONF), recently expanded its scope to explore the development and standardization of northbound APIs to enable development of network applications that run on OpenFlow networks. With HP and IBM now both offering OpenFlow controllers with Web-based APIs, the industry is poised to "create an ecosystem around the controller and develop the applications for automation and orchestration," said Bob Laliberte, senior analyst with Milford, Mass.-based Enterprise Strategy Group Inc. "Today there is no standardized format. They're working on it at the ONF, but until that time, you've got an opportunity to build out the ecosystem and provide value, which would be difficult to do if you don't have that controller.
"And enterprises are looking for solutions," he added. "They don't have a lot of developers and programmers on site to develop applications. They're looking for network providers to come to them with their own solutions, being able to certify end-to-end application."
IBM gets serious about networking with OpenFlow controller
The OpenFlow controller is another indication of IBM's return to the networking industry. The company first staked a claim to the server edge of the network when it acquired Blade Network Technologies two years ago.
"They've been strong supporters of SDN and OpenFlow in terms of standards activity. Now this demonstrates their intent to re-enter [the] market as a serious networking vendor," said Joe Skorupa, vice president and distinguished analyst for Stamford, Conn.-based Gartner Inc.
IBM has quietly used interest around software defined networking (SDN) and virtual network overlays as a means of establishing a larger footprint in networking. In January it introduced a top-of-rack OpenFlow switch, the RackSwitch G8264, along with a co-marketing and sales alliance with NEC's ProgrammableFlow controller. IBM has also introduced its own distributed virtual switch -- the 5000v -- and a virtual network overlay technology it calls Distributed Overlay Virtual Ethernet.
But IBM's growing network portfolio is still missing a core switch, for which it relies on partners like Brocade and Juniper Networks to provide. "[IBM's] current switches can be put together into a mesh that can support hundreds of servers in a scalable way, but with thousands of servers it gets messy. So a core switch would be necessary to maximize their opportunity," Skorupa said. By leveraging the ability of OpenFlow-based SDN to simplify network architectures, IBM could easily develop core switching products, he said.
"Entering the core switch market is dramatically different than it was before OpenFlow and SDN, particularly in the data center," Skorupa said. "If you're inside the data center, you probably don't need BGP 4 with 20 million routes and you don't need OSPF. You don’t need the very complicated software that was required in the pre-SDN world. With some of the merchant semiconductor capabilities and the ODM [original design manufacturer] houses that have sprung up, IBM could get a core switch with an OpenFlow stack on it for a small number of tens of millions of dollars."
Of course, IBM could also acquire a company that already has, or is close to having, an OpenFlow-enabled core switch.
IBM declined to discuss the future of its networking portfolio, saying that its partners can help fill the gaps in its product line. However, comments from IBM's Raphael suggested that the company does recognize a possible path beyond the server edge.
"Network vendors have been providing a really specialized set of hardware that is expensive and power-hungry, with millions of lines of code to support features. You've got thousands of RFCs that have to be supported and a lot of vertical integration with a lot of features baked into the infrastructure, where customers are very dependent on the vendors to come out with new features," Raphael said.
"So you have a bag of protocols to solve problems, and they are very complicated and CLI-driven. The administration is very complex because in order to achieve certain networking goals you have to manage down to the ACL, the VLANs, and do all kinds of traffic engineering. With IBM PNC, you get to this new paradigm where there's an opportunity to create a new, programmable data center. Here we can apply abstractions to the interfaces on the network to simplify the way these services are deployed."
IBM's OpenFlow controller isn't cheap
To engineers who have been experimenting with open source (i.e. free) OpenFlow controllers like Floodlight and NOX, IBM's PNC might present some sticker shock. One IBM OpenFlow controller license bundled with a license for one controlled switch is $92,000. Customers deploying the controller in a redundant pair as IBM suggests would have to make a $184,000 investment. To control more than one switch, customers will have to purchase additional licenses that cost at least $1,700 per device. That investment seems steep to Andre Kindness, senior analyst with Cambridge, Mass.-based Forrester Research Inc. First a company needs to buy OpenFlow switches. Then it needs to buy a license for the control plane.
"For a deployment of 10 switches, that's more than $10,000 in additional costs per switch," Kindness said.
Let us know what you think about the story; email: Shamus McGillicuddy, News Director.