Utility based framework for optimal network measurement

TY - JOUR

T1 - Utility based framework for optimal network measurement

AU - Parker, Ben M.

AU - Gilmour, Steven G.

AU - Schormans, John

PY - 2015/1

Y1 - 2015/1

N2 - Packet level measurement is now routinely used to evaluate the loss and delay performance of broadband networks. In active measurement, probe packets provide samples of the loss and delay and from these samples the performance of the traffic as a whole can be deduced. However this is prone to errors: inaccuracy due to taking insufficient samples, self-interference due to injecting too many probe packets, and possible sample-correlation induced bias. In this paper we consider the optimisation of probing rate by treating all measurements as numerical experiments which can be optimally designed by using the statistical principles of design of experiments. We develop an analytical technique that quantifies an overall utility function associated with: (i) the disruption caused per probe packet, (ii) the bias and (iii) the variance as a function of the probing (sampling) rate. Our numerical results show that the optimal probing rate depends strongly on what parameter the network engineer seeks to measure.

AB - Packet level measurement is now routinely used to evaluate the loss and delay performance of broadband networks. In active measurement, probe packets provide samples of the loss and delay and from these samples the performance of the traffic as a whole can be deduced. However this is prone to errors: inaccuracy due to taking insufficient samples, self-interference due to injecting too many probe packets, and possible sample-correlation induced bias. In this paper we consider the optimisation of probing rate by treating all measurements as numerical experiments which can be optimally designed by using the statistical principles of design of experiments. We develop an analytical technique that quantifies an overall utility function associated with: (i) the disruption caused per probe packet, (ii) the bias and (iii) the variance as a function of the probing (sampling) rate. Our numerical results show that the optimal probing rate depends strongly on what parameter the network engineer seeks to measure.

UR - http://www.scopus.com/inward/record.url?scp=84946558851&partnerID=8YFLogxK

U2 - 10.1049/iet-net.2013.0014

DO - 10.1049/iet-net.2013.0014

M3 - Article

AN - SCOPUS:84946558851

SN - 2047-4954

VL - 4

SP - 10

EP - 20

JO - IET Networks

JF - IET Networks

IS - 1

ER -