RAP-rate adaptive polling for network management applications

We present a new method for monitoring networks called Rate Adaptive Polling (RAP). The objective of RAP is to monitor network nodes at a sustained rate, and yet be non-intrusive to the performance of user applications in periods of congestion or near-congestion. RAP assumes a specific default polling rate for an application. Polls are issued by the manager at this rate. Feedback about active nodes and congestion is derived entirely through the responses to these polls. Future polling rate is adapted based on a congestion metric and timeout thresholds. RAP overcomes a few key limitations of existing polling engines used by popular network management platforms (including COP-N used by HP OV node manager). Specifically RAP detects the onset and abatement of congestion with a finer granularity. RAP also distinguishes between inactive nodes and congestion, unlike other algorithms. This ability along with its smoother backoff and increments helps improve its polling throughput while avoiding oscillations. Incorporating RAP within existing platforms is simple. No changes to the protocols or MIBs are required. We have implemented a RAP-based poller to monitor host-level multicast traffic on the Bell Labs Research intranet.