Is SDN really dead !? - Tech | 5G, SDN/NFV & Edge Compute

Breaking

Monday, July 8, 2019

Is SDN really dead !?

Is SDN really dead? Some would argue that SDN is dead. Some would argue that SDN is still alive. And the real answer is dependant on how you define SDN and what do you really mean by SDN.



A decade ago, when SDN technology was in infancy, people believed that it is going to bring ground-breaking architectural changes to the network. SDN architecture introduced three distinct layers in the network - data plane layer with all of the network elements, control plane layer with the SDN Controller and the application layer which makes the network, programmable. Many people believed that SDN Controller and SDN Applications would replace the several dozen routing protocols in the network. SDN, when it was introduced had three distinct objectives:

  • Separate the control plane from the data plane
  • Separate the hardware from the software (i.e., ability to run any vendor software on any vendor hardware)
  • Make the network functions programmable through APIs

Open Networking Foundation (ONF) championed the SDN initiatives and introduced the OpenFlow protocol. However, OpenFlow protocol didn't go past version 1.5.1. I would attribute SDN's failures to the path taken by ONF in realizing the SDN vision. Most of their projects were focused on OpenFlow and they made everyone to believe that OpenFlow is going to be the one stop solution for all of the problems in the network today. They tried to overload OpenFlow too much - for example, trying to do MPLS label distribution using OpenFlow extensions. Eventually, OpenFlow didn't scale beyond the datacenter needs. Established network equipment vendors also had vested interests to kill SDN. So, the OpenFlow momentum vanished in no time.

There are two significant contributions from the SDN community to the networking industry:

  1. The first contribution is Open Daylight SDN Controller. There were at least half-a-dozen SDN controllers that got introduced into the market during the initial days of SDN. To name a few - NOX, POX, Floodlight, ONOS, Open Daylight and Ryu. However the industry soon converged on Open Daylight, the most popular open source controller. The Open Daylight community has made 10 releases so far and has 50+ vendors supporting the community. Many vendors such as Cisco, Ericsson and Lumina are selling Open Daylight based SDN controllers today. Today, the community version of Open Daylight is bloated - as several vendors dumped proprietary/legacy APIs to the community and made Open Daylight a spaghetti. Despite the limitations, Open Daylight helps enterprises and service providers to simplify the management of networks - both physical as well as virtual networks.
  2. The second contribution is SD-WAN. Service providers invest in new technologies when those technologies either save costs (CAPEX/OPEX reduction) or help them generate newer revenue streams. SDN could potentially save a lot of money for service providers - however, the technology wasn't ready and it required service providers to rip/replace their existing network. The only SDN based service that helped service providers to monetize and generate new revenue was Software Defined WAN (SD-WAN). It is the only use case that garnered momentum in the industry. According to Global Market Insights, SD-WAN market is expected to hit $17 billion by 2025. There are at least two dozen SD-WAN implementations, mostly from SD-WAN startups. SD-WAN momentum is real and it would continue to co-exist with MPLS in the shorter-term.  
Beyond these two contributions, SDN didn't make a huge impact on the network or network architecture. 

The network that exists today was built over the last several decades. It is so complex that we cannot rip and replace the network in a short time. In another 10 years, most networks would be software-defined; most networks would be programmable; most networks would be virtualized. So, SDN as a concept may not be dead. If you think of SDN as OpenFlow, "Yes" it is dead. However, if you think of SDN as "improved programmability of network elements" and "reduced dependancy of networking software with proprietary hardware", SDN is going to grow in scale. As Steve Jobs would say, "You can't connect the dots looking forward; you can only connect them looking backwards". We've to wait and watch how the industry embraces SDN.

No comments:

Post a Comment