Page 68 - ITUJournal Future and evolving technologies Volume 2 (2021), Issue 1

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

arbitrary switches (where is the number of controllers 5. IMPLEMENTATION OF MATHEMATICAL
to place)) areselected as the potential controller locations MODELLING
(Instruction 3). This is followed by the association of each
switch to the closest controller (Instructions 4‑6). While This section explains our implementation for solving
the cost of con iguration (the overall propagation latency) the controller placement problem using the algorithms
decreases, the controller location and switch are described in sections 4.2 and 4.3. These algorithms

swapped (Instructions 7‑9), and each switch is reassigned are implemented in MATLAB 2018b. The primary
to their closest controller location (Instructions 4‑6). objective is to establish the number of controllers for
the achievement of minimum propagation latency and to
Algorithm 4 Partition Around Medoids (PAM) determine the best locations to place these controllers
Require: ( , ), , , in a WAN topology. The results from our simulation
, experiments are also presented in this section.
1: Compute shortest path matrix using Johnson’s
algorithm 5.1 Topologies
2: ←
3: ( ∈[1.. ]) ← randomly select objects from ( , , ) To maintain realism, our proposed solution is applied
on a real‑world WAN called South African Research
Network (SANReN), operated by CSIR’S Next Generation
4: for ∈ ( , ) do

5: Compute similarity score of with each ( ∈[1.. ]) Enterprises and Institutions (NGEI) cluster [50]. The

using reason for choosing the SANReN network was so that we
6: Associate to the most similar could demonstrate our proposed solution on an emerging

7: end for market use case. However, it may be noted that our
8: for do solutionistopology‑agnosticandcaneasilybeusedtotest

9: ← ( , ) other networks of different con igurations and sizes.

10: end for The SANReN network constitutes a core national
11: if ≤ 0 then backbone, with each Point of presence (PoP) connecting
12: ⇆ a metropolitan network. This work only focuses on

Go back to step 4 the PoP‑level instead of the router‑level view of the
13: else SANReN network. This is because the router‑level view
14: for ∈ ( , ) do is proprietary and not publicly available. Moreover,

15: Compute similarity score of with each the PoP‑level view has been deemed more useful [51]

( ∈[1.. ]) for several points: it provides a larger scale view of
16: Assign to the most similar network links, which are most interesting for network

17: end for optimization; it shows end users where they can connect
18: end if to the network and it’s the level where resiliency and
19: return
redundancy are critical. The PoP‑level geographical map
of the SANReN topology comprises 7 nodes and 7 ibre
If an increase in con iguration cost is detected, the swap is links con igured in a ring topology. The data set of this
undone and the optimal controller locations that optimize topology was downloaded from a repository called The
QoS are found (Instructions 12‑18). Two QoS parameters Internet Topology Zoo [52]. The format of the data set
are considered in our solution, that is the average is in Geography Markup Language (GML) and includes
propagation latency (which is the overall propagation geographic locations (longitude, latitude) of nodes and
latency) and the worst‑case propagation latency (which topological con iguration of the SANReN network.
is the maximum network latency). Eq. (6) and (7) de ine
′
how these parameters are de ined, where ( ) is the 5.2 Hardware and software used for
′
average latency, ( ) is the worst‑case latency, ( , ) modelling
is the shortest distance from the switch (node ∈ ) to
the controller (node ∈ ), = | | denotes the number All the experiments have been executed under an Ubuntu
8
of nodes and 2 × 10 is the speed of light in ibre. Desktop 16.04 LTS‑64 bit on a PC with the following
speci ication: Intel(R) Core(TM) i7‑5600U CPU, with 4
cores (8 threads), a clock speed of 2.60 GHz, RAM amount
1 of 8 GB and a storage capacity of 250 GB.
′
( ) = min ( , ) (6)
8
(2 10 ) ∈ ′
∈
5.3 Flowchart of proposed solution
The lowchart depicted in Fig. 1 summarizes the steps
in our proposed controller placement solution. First,
′
( ) = max min ( , ) (7)
∈ ∈ ′ network graph modelling is used to model the network

52 © International Telecommunication Union, 2021