(This article was written by Eric Carmès, 6WIND's Founder and CEO, for the Multicore Packet Processing Forum.)
Last July, Nicira’s acquisition by VMware shook (should I say shocked?) the networking industry. It was like a nostalgic retro move back to the good old Internet times. At least now everybody has heard about SDN and OpenFlow, although it’s unlikely that many people except a few visionaries can explain in a few words what SDN and OpenFlow really bring. Network virtualization, simplified network architecture, networking CAPEX and OPEX reduction, Network-as-a Service… all are nice concepts but not so easy to explain and articulate.
In 2012, OpenFlow standardization and education activities remained very strong within the Open Networking Foundation while new concepts like NFV (Network Function Virtualization) emerged to complement the OpenFlow architecture. However, SDN deployments are still very limited and at an early stage. Most of these SDN deployments don’t use OpenFlow. Another piece of technology, the virtual switch, was the subject of interesting discussions as it will likely play a pivotal role in SDN architectures.
Last year, a large number of SDN “me-too” software start-up companies emerged (should we suspect a Nicira effect?) and it’s quite difficult to understand their unique differentiation in the market. All these companies claim they have developed a smarter SDN architecture, typically including an optimized controller. At the same time, incumbent networking players explain that they fully support the SDN initiative… while reusing their existing protocols and promoting their own implementation of a virtual switch. Others say they have been designing and developing SDN-based architectures for years, arguing that SDN and OpenFlow are just marketing buzz.
So, will 2013 be finally the year of SDN? Maybe, but it will require both technical and business clarifications.
OpenFlow has been designed to be simple but is its simplicity compatible with complex network services? Will OpenFlow only address the operation and configuration of Layer2-3 hardware and software switches, or will it go beyond that? How will OpenFlow, the virtual switch, and NFV coexist to provide a complete SDN-based set of network services? Are we sure that the end solution will be based on open standards?
Beyond architecture design clarifications, emerging technologies always need to have a clear business justification to be widely adopted. How much will an OpenFlow architecture cost, including the switches and the controller, compared to a legacy solution? How much cost needs to be added to provide all the required Layer 4-7 services in a SDN-based data center? What about the OPEX savings? The industry is obviously still waiting for compelling business cases.
At 6WIND, we strongly believe SDN is going to transform the network industry and I explained why in a previous post. Having SDN architectures based on hardware Layer 2-3 switches, a high performance software-based data plane optimized for a standard server with a virtual switch as well as a distributed control plane represents a compelling approach to an agile, affordable networking architecture for data centers. It provides a solution for:
- Deploying a larger number of networked virtual machines for each server, beyond the top of rack, thanks to improved networking software performance and rich network services.
- Designing virtualized networking software appliances on standard server platforms for a fraction of the price of appliances based on custom hardware, while keeping power consumption under control.
2013 will likely tell us if SDN architectures will be based on open standards like OpenFlow or on proprietary architectures. The answer will be driven by business cases with a clear return on investment. 6WIND will contribute to the definition of these business cases, leveraging its high performance 6WINDGate™ data plane solution.