JOINT OPTIMIZATION

Improving system development through joint optimization



Improving system development through joint optimization

Introduction

This paper will be focusing on improving the networking systems in the organization (Salhi, 2010). One of the highly desired features in network monitoring is to minimize the cost of deploying and maintaining monitoring devices, and reduce communications between the monitoring devices and the Network Operations Center (NOC); while preventing monitoring flows from interfering with real traffic flow. These features could be achieved by minimizing the number of monitors that are to be deployed, minimizing the number of monitored paths; respectively. However, we argue that there is interplay between these objectives. Indeed, reducing the number of monitoring devices and the number of monitoring flows results in monitoring long paths that are quite likely to overlap, which increases redundant measurements. Our previous work illustrated this conflict aspect through ILP formulations. We proposed a monitoring cost model that includes a monitor location cost and an anomaly detection cost. Simulation results demonstrated that a joint optimization of the two costs reduces efficiently their trade-off. Typically, to overcome scalability issues, the set of candidate paths that are to be monitored is restrained to a small subset. Unfortunately, none of the related works investigated the impact of this restriction on the quality of the monitoring solution neither did they specify how to choose the set of candidate paths. In this paper, we further investigate these issues. In this paper, we proposed novel greedy algorithms for joint optimization of monitor location and network anomaly detection. Our main goal was to come up with large-scale heuristics that reduce the overall monitoring cost (Lahoud, 2010).

Discussion

Acknowledging the efficiency of the joint optimization model to balance the trade-off between the multiple minimization objectives, we keep on considering this technique to devise novel greedy algorithms (Zhao, 2009). The aim is to come up with large-scale heuristics that apply for large networks and achieve good quality solutions. The initial algorithm that we call exhaustive greedy algorithm starts by selecting a pair of monitor locations that maximizes the number of covered links; and then it explores all paths between the selected monitors, in order to choose a set of paths that maximizes the number of links it can cover and minimizes redundant measurements. Additional monitors and paths are selected iteratively in order to reduce the number of overlaps among paths that are to be monitored, thereby reducing the cost of the solution. The second algorithm that we call selective greedy algorithm is based on a heuristic that minimizes drastically the number of explored paths (Salhi, 2010). The underlying idea is to choose a set of non-overlapping paths that maximizes link coverage, and then select additional paths and potentially additional monitors to cover the remaining links. The paths of this set should be long enough to increase the number of covered links. However, we do not require them to be the longest paths of the network. This relaxation avoids exploring all paths between selected ...