In-Network Monitoring

Abstract

Monitoring, i.e., the process of acquiring state information from a network or networked system, is fundamental to system operation. In traditional network and systems management, monitoring is performed on a per-device basis, whereby a centralized management entity polls the devices in its domain for information, which is then analyzed and acted upon. In this chapter, we describe several monitoring algorithms that utilize a new monitoring paradigm called In-network Monitoring. This paradigm is designed to address the above shortcomings, and we demonstrate how it can be applied to managing highly dynamic networked systems. The main idea of In-network Monitoring is to introduce a small management entity inside each network device, which, in addition to monitoring local parameters, can also perform limited management functions and communicate with peering entities in its proximity. The collection of these entities creates a monitoring layer inside the network, which can perform monitoring and control tasks without involving the centralized entity. We demonstrate how In-network monitoring can help building better and more efficient systems. We start with a general description of network monitoring techniques, and then describe two specific cases in which this paradigm generates provably efficient solutions. The first one is in the area of traffic engineering, where there is a need to monitor the aggregated delay of packets along a given network path. The second case deals with the problem of monitoring general aggregated values over the network, with emphasis on computing the values in a distributed way inside the monitoring layer. All together, we believe that this new paradigm presents a promising direction to address the challenges of cost-effective management of future networked systems.