The Available Bandwidth Measurement Based On Hidden Markov Model

With the rapid development of Internet, new network applications constantly appear inpeople’s lives, the rapid growth of online shopping, online payment, making the network qualityof service is facing many challenges. Available bandwidth, as an important indicator of networkperformance, can be thought to provide us with the reference on monitoring network behavior,improving the efficiency of data transmission and network planning. How fast, accurate access tothe available bandwidth in the path has become the difficulty of the study and has importantpractical significance.This paper first introduces the bandwidth measurement infrastructure, including the basicconcepts of bandwidth measurement, active measurement and passive measurement techniques,as well as popular bandwidth measurement tool, discussing the factors that impact the availablebandwidth measurement results. These factors include the errors caused by end-to-end hosts androuters, followed by the burst background traffic in the path will also affect the measurementresults. Two typical end-to-end path available bandwidth measurements based on the packet ratemodel (PRM) and the packet gap model (PGM) are elaborated. The advantages anddisadvantages of these two methods are also analyzed. So a spruce improved method ofmeasuring available bandwidth (Wspruce) is proposed based on the feature of spruce.The method is based on Hidden Markov Model (HMM). In an end-to-end path, availablebandwidth of each moment is changing, high and low. These states can not be directly observed,but the observed distribution of variable probes can be used to indirectly identify the hidden statesequence. The states in the HMM is correspondingly unobservable, but they can be estimated byobserving the variables. So the states of the available bandwidth can be regard as the states in theHMM, the variables of probe packets distribution as the observed variables, then HMM can beestablished for the available bandwidth. By using HMM series prediction features, availablebandwidth can be made more accurate analysis. Actual measurements show that the method forestimating the available bandwidth measurement is faster, lower overhead.