Page 73 - ITUJournal Future and evolving technologies Volume 2 (2021), Issue 1

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

7.2 Methodology do this we generate an ICMP packet using the ping utility
for each pair. This is followed by computation of the
This work constitutes two independent experiments. ICMP pinging results to obtain the total average latency
The irst experiment is carried out with the intention (round‑trip time) from the node to all other nodes in the
to address the controller placement problem leveraging network. Thisstepis repeatedforall nodesintheSANReN
emulation. The irst experiment analyses two scenarios: topology. To ensure valid and reliable results, we repeat
(1) when only one controller is used and (2) when two the above procedure 5 times under a soft idle timeout for
controllers are used. The second experiment presents the controller entry of 5 seconds (the soft idle timeout
different approaches through which signalling overhead de ines the expiry time of a controller low rule when
on the control channel can be reduced, in consideration there is no low activity) and compute the average results.
of control‑plane resiliency. The soft idle timeout is set to ensure generation of control
traf ic upon pinging reiterations.
7.2.1 Controller placement
For the case of two controllers (see Fig. 10), the network
On example of one controller case, Fig. 9 summarizes our is partitioned into two smaller administrative domains,
approach in a low chart (where is the total potential namely cluster one and cluster two, each supervised by
controller placement locations, i.e. the total number of a dedicated ONOS instance. The parameters 1 and 2
nodes in a given topology). For the SANReN network, denote the total number of switches in cluster one and
is 7, meaning there are a total of 7 potential controller two, respectively. After executing the mastership module,
placement locations in the network. the partition results are as follows: The irst ONOS
instance ( 1) is assigned three switch nodes in region
The following procedure (outlined in Fig. 9) is used for Pretoria, Bloemfontein and Durban, while the other ONOS
each node to determine average latency: To ind optimal instance constitutes switches located in Johannesburg,
controller locations, irst we install the ONOS controller Cape Town, East London and Port Elizabeth. In order
in the same geographic location as the irst OpenFlow to optimize the placement of these two controllers, an
switch node (using the Harvesine great circle approach exhaustive search is carried out by iterating through all
and the Linux TC utility). The next step is to trigger possible combinations (within the limits de ined by each
a packet‑In message to the controller. This is done by controller domain). In other words, 1 is placed at
generating traf ic lows between all pairs, i.e. between different regions within its domain. For each placement
this node and all other nodes in the SANReN topology. To of 1, 2 is then placed at different regions within its

Fig. 9 – Flow chart of proposed method for one controller[56].

© International Telecommunication Union, 2021 57