A Case Study of the Accuracy of SNMP Measurements

For some time it has been known that the standard method for collecting link-traffic measurements in IP networks — the Simple Network Management Protocol or SNMP— is flawed. It has oft been noted that SNMP is subject to missing data, and that its measurements contain errors. However, very little work has been aimed at assessing the magnitude of these errors. This paper develops a simple, easily applicable technique for measuring SNMP errors, and uses it in a case study to assess errors in a common SNMP collection tool. The results indicate that most link-load measurement errors are relatively small, but the distribution has a heavy-tail, and that a few measurement errors can be as large as the measurements themselves. The approach also allows us to go some way towards explaining the cause of the errors.