E-Locus - Institutional Repository of the University of Crete

Home Search

Results - Details

Search command : Author="Μουχτάρης" And Author="Αθανάσιος"

Current Record: 7 of 29

[Add to Basket]

Identifier

000413378

Title

Farcast : improving forecasting via SDN

Alternative Title

Farcast: Βελτίωση του Forecasting μέσω των SDN

Author

Μπαμιεδάκης – Πανανός, Μιχαήλ Ν.

Thesis advisor

Δημητρόπουλος, Ξενοφώντας

Reviewer

Παπαδοπούλη, Μαρία
Μούχταρης, Αθανάσιος

Abstract

Forecasting in computer networks is the process of anticipating future traffic demands based on present and past data and by analysis of trends. Network providers use forecasting in order to examine whether the traffic is routed efficiently, also they use it for bandwidth allocation and congestion control. In this work, we propose an approach, which we call Farcast, for improving the accuracy of network traffic load forecasting using Software Defined Networking (SDN) principles. In Farcast, the traffic load predictions for monitored links are sent to an SDN controller. The SDN controller has a global view of the network and monitors the network state in time frames called inform intervals. In these intervals, the SDN controller receives messages from network switches, about the current traffic flows traversing the network. Then, the SDN controller assesses whether the predictions will be accurate or not, based on the current traffic demands. If there are imminent inaccurate predictions, the controller ensures that the prediction error is reduced, by re-routing, when possible, the flows in the network accordingly. Farcast was implemented in a simulated SDN environment. The experiments simulate traffic bursts that are difficult to predict with traditional forecasting approaches. We compare the performance of the AutoRegressive Integrated Moving Average (ARIMA) forecasting method with Farcast. The results show that Farcast reduces the Mean Absolute Percentage Error (MAPE) of the traffic load predictions, by up to one order of magnitude, when the goal is to satisfy the predictions for a single link in the network. Moreover, when adjusting the load of multiple monitored links concurrently, Farcast reduces the MAPE by up to 50%. Our approach can be deployed independently of the machine learning algorithm used for the predictions. For example, such a approach would be useful in the 40-Gigabit-capable Passive Optical Network (XG-PON) standard. The PON is composed of Optical Netwok Units (ONUs) and an Optical Line Terminator (OLT). The ONUs predict the future traffic load and send bandwidth requests to the OLT, based on their predictions. The OLT performs the dynamic bandwidth allocation based on the ONUs’ requests. In such schemes, the dependability on forecasting is critical, as wrong predictions can lead to underutilization of system resources or traffic congestion.

Language

English

Subject

ARIMA

Computer networks

OMNET

Issue date

2017-11-24

Collection

School/Department--School of Sciences and Engineering--Department of Computer Science--Post-graduate theses

Type of Work--Post-graduate theses

Permanent Link