DocumentCode
1430484
Title
Adding Percentiles of Erlangian Distributions
Author
Anjum, Bushra ; Perros, Harry
Author_Institution
Comput. Sci. Dept., North Carolina State Univ., Raleigh, NC, USA
Volume
15
Issue
3
fYear
2011
fDate
3/1/2011 12:00:00 AM
Firstpage
346
Lastpage
348
Abstract
In networking, enterprise computing and many other areas the issue of adding percentiles of a performance metric, such as the response time, arises regularly. Percentiles cannot be added using the arithmetic sum, and surprisingly there are no known formulae that permit us to do so correctly. In this paper, we obtain an exact analytical expression for adding percentiles of random variables which can be represented by a series of generalized exponential stages (e.g., Erlang, hypoexponential and two-stage Coxian). We demonstrate the applicability of our results by an example in which we use our expressions in the Dijkstra´s algorithm to calculate the shortest ´percentile delay´ path.
Keywords
business communication; network theory (graphs); performance evaluation; random processes; Dijkstra´s algorithm; Erlangian distribution; enterprise computing; percentile delay; performance metric; random variable; Delay; Jitter; Quality of service; Random variables; Routing; Time factors; Adding percentiles; Erlang; hypoexponential; quality of service; shortest path calculation; two-stage Coxian;
fLanguage
English
Journal_Title
Communications Letters, IEEE
Publisher
ieee
ISSN
1089-7798
Type
jour
DOI
10.1109/LCOMM.2011.011011.102143
Filename
5692887
Link To Document