TY  - CONF
EP  - 583
A1  - Said, A.M.
A1  - Hasbullah, H.
A1  - Dahab, A.Y.
UR  - https://www.scopus.com/inward/record.uri?eid=2-s2.0-80053147628&doi=10.1109%2fICCSN.2011.6014960&partnerID=40&md5=95e3b567fe5ac5a11b1d7f9e53ff6591
SN  - 9781612844855
Y1  - 2011///
TI  - R-largest order statistics for the prediction of bursts and serious deteriorations in network traffic
SP  - 579
ID  - scholars1893
KW  - Analysis and modeling; Block maxima; bursts; Extreme value theory; Heavy-tails; Internet traffic; Network traffic; Order statistics; Peaks over threshold; Prediction problem; Quality of service metrics; Self-similarities; Service Level Agreements; Service provider
KW  -  Communication; Deterioration; Forecasting; Networks (circuits); Quality of service; Telecommunication traffic
KW  -  Wind effects
N1  - cited By 1; Conference of 2011 IEEE 3rd International Conference on Communication Software and Networks, ICCSN 2011 ; Conference Date: 27 May 2011 Through 29 May 2011; Conference Code:86671
N2  - Predicting bursts and serious deteriorations in Internet traffic is important. It enables service providers and users to define robust quality of service metrics to be negotiated in service level agreements (SLA). Traffic exhibits the heavy tail property for which extreme value theory is the perfect setting for the analysis and modeling. Traditionally, methods from EVT, such as block maxima and peaks over threshold were applied, each treating a different aspect of the prediction problem. In this work, the r-largest order statistics method is applied to the problem. This method is an improvement over the block maxima method and makes more efficient use of the available data by selecting the r largest values from each block to model. As expected, the quality of estimation increased with the use of this method; however, the fit diagnostics cast some doubt about the applicability of the model, possibly due to the dependence structure in the data. © 2011 IEEE.
AV  - none
CY  - Xi'an
ER  -