State Acquisition in Computer Networks

We establish that State Acquisition should be per-formed in networks at a rate which is consistent with the rate-of-change of the element or service being observed. We demonstrate that many existing monitoring and service-level prediction tools do not acquire network state in an appropriate manner. To address this challenge: (1) we define the rate-of-change of different applications; (2) we use methods for analysis of unevenly spaced time series, specifically, time series arising from video and voice applications, to estimate the rate-of-change of these services; and finally, (3) we demonstrate how to acquire network state accurately for a number of real-world traces using Greedy Acquisition. The accuracy of State Acquisition is improved when it is performed at a rate which is consistent with its rate-of-change. An improvement in representation accuracy of one order of magnitude is achieved for voice and video streaming applications; this improvement does not incur any additional bandwidth or storage cost.