Data sources in proactive network management

As networks are extremely heterogeneous, and application layer requires higher level o f flexibility in network performance and resource allocation, proactive management is becoming an important research target. Main purpose of proactive management is to detect network performance anomaly before its occurrence and undertake steps to rectify the conditions that lead to the anomaly. Since detection of anomalies is the key point of proactive management, continuous data about network performance are required. Conventionally, to obtain performance data one would use SNMP protocol to poll MIB agents at networking devices at regular intervals, and then process the data offline. For large management domains offline processing either takes long time when thorough, or is not reliable when processing is made selective. To solve the above problem with online processing, we propose to use performance data obtained by end-to-end probing. In our study, we use neural network to predict anomalies. Comparison of predictions made solely based on SNMP polls with those that use end-to-end probing prove the validity of our proposal. End-to-end performance data offers clearer patterns, and best error rate of predictions around 4-6%, which is one forth of predictions based on SNMP polls. In our study we use special probes with packets of two different sizes in order to obtain multiple performance data from a single probe.