Page 61 - ITUJournal Future and evolving technologies Volume 2 (2021), Issue 1
P. 61

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 1







          CONTROLLER PLACEMENT OPTIMIZATION FOR SOFTWARE DEFINED WIDE AREA NETWORKS
                                                       (SDWAN)

                                                                       2
                                    Lusani Mamushiane 1,2 , Joyce Mwangama , Albert Lysko 1,2
                     1 Council for Scienti ic and Industrial Research (CSIR), University of Cape Town, South Africa
                                                                   2
                               NOTE: Corresponding author: Lusani Mamushiane, Lravhuanzwo@csir.co.za


          Abstract – Software De ined Networking (SDN) has emerged as a promising solution to revolutionize network deployment,
          operations and economic growth. This paradigm aims to address management and con iguration complexities in legacy
          networks so as to reduce the total cost associated with deploying and running telecommunication infrastructures. At the heart
          of SDN is a controller which oversees orchestration of resources. An important problem that must be addressed during the
          initial design of an SDN‑based network deployment is to  ind the optimal number of controllers and their locations, to achieve
          desired operational ef iciency. This problem constitutes competing objectives such as latency, load balancing, and reliability.
          We apply Silhouette Analysis, Gap Statistics and the Partition Around Medoids (PAM) algorithms and, unlike previous work,
          we add a new method for solving the controller placement problem using an emulation orchestration platform. Our approach
          aims to optimize controller‑to‑node latency, alleviate control‑plane signalling overhead and ensure control‑plane resiliency.
          Our results for South African national research network (SANReN) reveal that deploying two controllers yields the lowest
          latency, reduces control‑plane signalling overhead and guarantees control‑plane resiliency. Our approach can be used by
          network operators as a guideline to start integrating SDN or plan a new SDN deployment, by helping them make quick
          automatic decisions regarding optimal controller placement.

          Keywords – Controller placement, Gap Statistics, Partition Around Medoids, Silhouette, Software De ined Networks,
          South African National Research and Education Network




          1.    INTRODUCTION                                   Software   ined Networking (SDN) has emerged


                                                               as a promising candidate to revolutionize future










          Over the past  decade, the use of information and    telecommunication  landscapes.    Contrary to the










          communication technology has reached the upper       traditional  network architecture where the control





          bounds of Internet penetration [1]. According to a Cisco   and data‑plane  of packet  processing devices are tightly





          White paper [2], Internet usage is anticipated to continue
                                                               coupled, SDN presents a paradigm shift in networking by

          on an upward trajectory in the foreseeable future. This


                                                               decoupling the control‑plane  logic from the underlying



          strong appetite for Internet access is causing a high









                                                               physical infrastructure [5]. The control‑plane  is then





          demand for bandwidth and putting    icant pressure





                                                               logically  centralized in an external  entity called  a



          on the existing  telecommunication  infrastructure.


                                                               controller and interacts with the physical infrastructure




          There is a consensus that  the current infrastructure   via its southbound interface.    By decoupling the








          will  not   ice to cater for these exploding  demands







                                                               control logic from the physical hardware, operators












          [3]. This is primarily attributed to the rigidity of the   can programme new   ic engineering policies (such










          legacy infrastructure, especially because of vendor   as bandwidth management,  security, protection and









          lock‑in (the use of proprietary silicon hardware) which   restoration policies) without  worrying about  the









            les innovation and makes it   icult to scale the   constraints of closed proprietary hardware and  irmware.







          network on the   ly. As a result of vendor lock‑in, the cost   Moreover,  the abstraction  of  lower level functionality








          associated with upgrading the infrastructure to cater   provided by SDN enables convergence of heterogeneous

          for the changing tr  ic patterns is very high, meaning   hardware thereby fostering a vendor‑neutral ecosystem.
















          adding new features ad hoc is virtually impossible [4].   In addition to enabling centralized network provisioning






          Therefore, network operators desiring new features to   and holistic network management,   SDN promises




          address their market needs end up beholden to a vendor’s     its such as security granularity  (by providing a














          upgrade timelines and costs. To cater for the increase   central point of control to holistically  and consistently







          in Internet demand, the infrastructure has to evolve   disseminate security information), savings in operational






          from its current monolithic nature to a vendor‑agnostic,   costs (by automating  network administrative tasks),






          programmable,  cost‑effective (in terms of deployment   savings in capital  expenditures (by capitalizing  on






          (CapEx) and operational costs (OpEx)) and more  lexible   commodity hardware) and cloud abstraction  (which is





          infrastructure.

                                                               critical to consolidate and facilitate  the management  of




                                                               massive data centers) [6]. According to [7] a huge portion
                                                               of operational expenditure is from costs related to the








                                             © International Telecommunication Union, 2021                    45
   56   57   58   59   60   61   62   63   64   65   66