Peer-to-Peer and Distributed System Measurement Request Patterns

Abstract

Measurement service has knowledge of a larger number of network
measurements than individual applications, it is in a position to determine when inference
can be used to reduce the total number of measurements required to satisfy a particular
demand from applications. To accomplish this, the service must quantify the measurement
load required to operate a network inference mechanism, and compare this load to that of
direct measurement of the requested properties. In this paper, we predict the network
traffic injected by inference mechanisms, and use this knowledge to replace direct
measurement traffic by inference when the cost of direct measurement exceeds that of
inference. After setting up an inference mechanism, continuing measurements typically
require O log(n) probes to estimate the properties of O(n2) paths