Page 72 - ITUJournal Future and evolving technologies Volume 2 (2021), Issue 1

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

Table 3 – Average ( ) and worst‑case ( ) latency for varying number of controllers

= 1 = 2 = 3 = 4
( ) 2.9 1.81 1.2 0.98
( ) 6.8 3.92 5 5.3
Johannesburg
Pretoria
Pretoria Durban
Names of locations for Durban East London Johannesburg East London
Port Elizabeth
Port Elizabeth
Pretoria
Cape Town
Port Elizabeth Cape Town
Names of locations for Bloemfontein Bloemfontein East London Bloemfontein
Durban
Port Elizabeth

SANReN, the same network we used in Section 5. It may
however be noted that our approach is generic and can be
used to optimize any other network.

7.1 Experimental setup

The experiment setup is as illustrated by Fig. 7 and Fig.
8 (captured from Miniedit). Node c0 and c1 are ONOS
SDN controller instances running on a dedicated remote
machine (with 8 CPUs, 16 GB RAM and 1 TB HDD and
no swap partition), and h0‑h6 are hosts attached to SDN
Open Virtual Switches (OVS 2.9.90) running OpenFlow
version 1.3. A built‑in application for reactive low
instantiation is activated to set the ONOS controller to
reactive operational mode. The red dash‑dotted lines
show connection (over WiFi) between switches and
controllers and the blue solid lines are links between Fig. 7 – Experiment setup with one ONOS controller
the switches. The switch‑to‑controller communication is
assumed to happen out‑of‑band. Since the links between
the switches are known to be ibre, where speed is
approximately the speed of light in ibre i.e. about 2 ×
8
10 / , we use the latency formula Eq. (9) to con igure
the link properties.
( )
( ) = (9)

The distances between nodes are calculated using the
Harvesine great circle approach and the actual GPS
coordinates of the nodes.
The data‑plane emulated on Mininet version 2.2.2 (with
default settings, for all experiments) is running on
a separate machine (with 8 CPUs, 16 GB RAM and
1 TB HDD). Each switch in the data‑plane has a Fig. 8 – Experiment setup with two self‑coordinating ONOS controllers
unique data path ID (DPID). The connection between
the control‑plane and data‑plane is via port 6633 of
the controller over a slow WiFi router. The control
link parameters are con igured using the Linux Traf ic
Control (TC) utility (installed on the machine used for
data‑plane emulation) under the assumption that the
optimal controller placement is co‑located with one of the
switches. The programming language used to develop the
software is Python 2.7.14.

56 © International Telecommunication Union, 2021