next up previous
Next: Conclusion Up: Network Management of Predictive Previous: Optimizing Management Polling with

Interaction between a Predictive Management System and a Predictive Mobile Network

There is an interesting interaction between the predictive management system and the predictive mobile network. A predictive mobile network such as the VNC proposal for RDRN [4] will have results cached in advance of use for many configuration parameters. These results should be part of the Management Information Base (MIB) for the mobile network and should include the predicted time of the event which requires the result, the value of the result, and the probability that the result will be within tolerance at that time. Thus there will be a triple associated with each predicted event: (time, value, probability). Network management protocols, e.g. SNMP [10] and CMIP [11], include the time as part of the PDU, however this time indicates the real time the poll occurred and should not be changed.

A predictive management system could simply use LPs to represent the predictive mobile processes as previously described, however, this is redundant since the mobile network itself has predicted events in advance as part of its own management and control system. Therefore, managing a predictive mobile network with a predictive network management system provides an interesting problem in trying to get the maximum benefit from both of these predictive systems.

Combining the two predictive systems in a low level manner, e.g. allowing the LPs to exchange messages with each other, raises questions about synchronization between the mobile network and the management station. However, the predicted mobile network results can be used as additional information to refine the management system results. The management system will have computed (time, value, probability) triples for each predicted result as well. The final result by the management system would then be an average of the times and values weighted by their respective probabilities. An additional weight may be added given the quality of either system. For example the network management system might be weighted higher because it has more knowledge about the entire network. Alternatively, the mobile network system may weighted higher because the mobile system may have better predictive capability for the detailed events concerning handoff. Thus the two systems do not directly interact with each other, but the final result is a combination of the results from both predictive systems. A more complex method of combining results from these two systems would involve a causal network such as the one described in [23].


next up previous
Next: Conclusion Up: Network Management of Predictive Previous: Optimizing Management Polling with

Steve Bush
Thu Feb 27 15:34:42 CST 1997