ACM Computing Surveys, December 2014.
present an overview of the major design trends and highlight key differences between them.
Active networking focussed on improving the data plane functionality of the network ... PlanetLab [Chun et al. 2003] feature the separation of traffic to different execution environments on the basis of packet headers
SDN efforts differed from active networking by focussing on problems of immediate import to network administrators
The application of these concepts in the context of wireless networks poses many challenges.
major design attempts at bringing SDN concepts to wireless networks, classified according to target networks, for example, WLAN or cellular
Much of the research in wireless SDN so far have focussed on IEEE 802.11 networks. Perhaps this has to do with the fact that network devices are not as closely tied to the architecture as it is in the case of cellular networks.
considerable interest on the part of cellular companies: offloading data traffic to WiFi networks whenever possible thus SDN efforts in WiFi and cellular are not entirely isolated from each other
In campus WiFi networks, network administration is mostly centralized. Commercial products from various companies … OpenFlow Wireless aims to be an open alternative …
Yap et al. [2010] envisage a future where the end user is free from worrying about the details as to which wireless network she is getting service from... a user roams freely between cellular and WiFi networks taking advantage of seamless handovers
FlowVisor [Sherwood et al. 2010] to delegate the control of different slices to different controllers. The FlowVisor is essentially a proxy that forwards OpenFlow messages from different slices to the appropriate controllers. While a FlowVisor achieves slicing of flows, it does not provide facilities for configuration of the network.
separating the control and data plane allows outsourcing of network configuration and management
Streaming applications like YouTube or Netflix can benefit greatly from a customized network slice.
not immediately clear how much of network management and control can be centralized in view of the fact that wireless channel conditions are variable
programmability of the PHY and MAC layers by attempting to define a software abstraction layer that hides the hardware details
simplify the implementation of high level enterprise WLAN services, such as authentication, authorization and accounting (AAA)
light virtual access points (LVAPs)
SoftRAN [Gudipati et al. 2013] — improving the design of the Radio Access Network (RAN). RAN … providing wide area connectivity to mobile devices.
a global view of power allocations across subcarriers at each base station enables us to make better power allocation decisions than what one could have hoped for using a distributed algorithm, as is done at present
latency considerations would dictate which functionality can be handed over to the controller ... Overall the goal is to have a network operating system running over the cellular infrastructure, there enabling various network management requirements to be written as application modules.
SoftRAN abstracts out all radio resources of a geographical area in a three dimensional grid of base station index, time and frequency slots. A geographical area is defined as a macro cell.
reduced reliability of the control channel
high variability of WSNs and its application specific nature … difficult to manage
proposed architecture - control plane is decoupled from the data plane that runs on the sensor nodes - centralized controller uses a customized version of OpenFlow to interact with the nodes - Luo et al. [2012]
extending OpenFlow to support WSN specific use cases, such as flow creation using sensor attributes
Compared to the data network, the control network is typically subject to better standards of reliability. Particularly in WSNs, as node and link failures are much more common, the design choice of sharing the network remains dubious. Other difficulties: ... minimize control overhead as communication is inherently costly
Tenet [Gnawali et al. 2006] decouples control from the sensor motes --- argue for a tiered architecture for WSNs. The lower tier consists of resource constrained motes while the upper tier contains fewer but more capable nodes called masters.
the Tenet design principle may be stated as:
“Multi-node data fusion functionality and multi-node application logic should be implemented only in the master tier. The cost and complexity of implementing this functionality in a fully distributed fashion on motes outweighs the performance benefits of doing so.”
written in software in the master tier, combine this functionality to achieve network objectives. These properties are characteristic of SDN.
SDN opens many axes in the network design space.
increased channel variability and the latency sensitive nature of key network parameters ... the larger question of how to make use of this new-found freedom optimally for different scenarios remains unanswered