Publications

Comprehensive Prediction Models of Control Traffic for SDN Controllers

2018 4th IEEE Conference on Network Softwarization and Workshops (NetSoft) IEEE, Montreal, QC, Canada, 2018, pp. 262-266, doi: 10.1109/NETSOFT.2018.8460111.

Bong-yeol Yu, Gyeongsik Yang, Chuck Yoo

Download PDF

Abstract

In SDN, as the control channel becomes a performance bottleneck, modeling the control channel traffic is important. Such a model is useful in predicting the control channel traffic for network provisioning. However, previously proposed models are quite limited in that they assume only the forwarding function of a specific controller for their models. To overcome the limitations, first, this paper analyzes the control traffic by seven functions (including forwarding function) of a controller. Then, we build a seven-function model to predict control channel usage and evaluate the prediction accuracy that achieves as high as 94%. Note that previous models did not have any quantitative evaluation. Our model is built with the Open Network Operating System (ONOS) controller and extended to Floodlight and POX controllers. We show that the extended model also achieves similar prediction accuracy (95%). Furthermore, we compare three controllers in terms of control channel usage through our model.

Paper summary

Explain control traffic by controller function

An SDN controller’s traffic can be estimated from the operations it performs and the network it manages.

The study derives control-traffic models for ONOS, Floodlight and POX using their default applications. It analyzes exchanged packets and separates functions such as connection setup, discovery, statistics and forwarding.

Build a model from the controller’s individual functions

Mathematical terms relate control-message counts and sizes to switches, ports, hosts, flows and hop distances. The model separates connection setup, discovery, statistics collection, forwarding and other controller functions, then combines their contributions. This also explains why controllers can produce different traffic volumes on the same topology: their default applications perform different amounts of monitoring and discovery work.

Network quantities used to estimate controller traffic.
Input groupQuantities
InfrastructureSwitches, ports and hosts
TrafficFlows and path hop counts
Controller behaviorSetup, discovery, statistics and forwarding functions

Compare predicted traffic with measured traffic

Mininet experiments cover three fat-tree sizes and three ISP-topology configurations, using ONOS, Floodlight and POX. The paper reports 94.8% average prediction accuracy across its evaluation. It also identifies discrepancies caused by temporary traffic such as neighbor discovery, bootstrap and DNS requests. The estimates describe the analyzed controllers and default applications; a controller with different applications or polling behavior needs corresponding model terms.

Prediction accuracy for ONOS, Floodlight and POX control-traffic models.View full-size figure
Control-traffic prediction accuracy for ONOS, Floodlight and POX. The three models achieve similar accuracy.

Key takeaways

  1. Control traffic reflects what the controller does. Separating discovery, monitoring and flow processing explains why topology alone cannot determine message volume.
  2. An interpretable model can connect network structure to traffic demand. Function-specific terms combine switch, host, flow and path quantities into the prediction.