Page 74 - ITUJournal Future and evolving technologies Volume 2 (2021), Issue 1
P. 74

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




















































                                        Fig. 10 – Flow chart of proposed method for two controllers.


          domain. For each set of placement, the average latency  of the switch). By partitioning the data‑plane into two
          is computed following the same procedure as described  clusters, the traf ic induced by data‑plane monitoring
          above.                                               (code‑named polling) is reduced.  Speci ically, after
                                                               clustering, the controller sends and receives monitoring
          7.2.2  Control‑plane signalling overhead and         data from just a fraction of data‑plane nodes. To balance
                 failover                                      the switch‑to‑controller placement, we activate the ONOS
                                                               mastership management module. This results in a more
          The centralized control scheme adopted by SDN puts   balanced monitoring load which we expect to further
          the control channel at risk of incurring very high   decrease control‑plane overhead.
          signalling  overhead  generated  during  data‑plane
          monitoring (e.g.  Stats‑Request and Stats‑Reply) and  To quantify the impact of switch‑to‑controller placement,
          reactive  low instantiation (such as packet‑In, packet‑Out  we generate variable traf ic between two virtual hosts
          and Flow‑Removed).  In order to manage this rapid    (the client connected to Johannesburg and the server
          in lux of traf ic on the control channel, the following  connected to Cape Town), a distance of 1399 km from
          procedure is used:  First we con igure a cluster of  each other. This is carried out using the Distributed
          two ONOS instances each managing a segment of the    Internet Traf ic Generator (D‑ITG) tool. The transport
          network. The cluster is con igured using the REST API  protocol is set to UDP and the number of packets
          of each separate ONOS. Upon data‑plane instantiation,  per second is varied from 50 000 to 200 000 in
          the switch‑to‑controller placement (in terms of the  increments of 50 000. The packet size is set to 512
          number of switches per cluster) is imbalanced. This is  bytes.  The link bandwidth was kept at 10 Mb/s.
          because switch‑to‑controller placement is based solely  The duration for the generation process is set to 5
          on best effort (meaning the controller that completes  minutes.  The key performance indicators are delay,
          the handshake with the switch  irst, gets mastership  jitter and packet loss all monitored at the server end.





          58                                 © International Telecommunication Union, 2021
   69   70   71   72   73   74   75   76   77   78   79