Markovian Model for Packet Loss in MPLS Networks

In this Paper, we worked on the modeling of packet loss within MPLS environment. The research exploited real data from a real network running MPLS as its core switching. The data were obtained from 6 nodes, each node participated with 6 hours of traffic segments used as data set, all nodes formed 36 hours as testing data. Each dataset was divided into one-hour segment, each segment loss and no-loss was represented as a binary series. From the 36 segments; 26 segments were found stationary. For Stationary segments;9 segments were modeled using Bernoulli model, 16 segments were modeled using 2-state Markov chain and 9 segments showed k-th order Markov chain tendencies with orders higher than 2. these segments were excluded from modeling due to the uniqueness of their conditions. 4 other segments were found non-stationary thus couldn't be modeled. Each segment stationarity was checked using average filter conditions with threshold of 0.05 and with confidence of 0.95 for lags selection. After modeling each segment independently; an average loss model for each node was estimated using its modeled segments. confidence of these models were 95%. The Model error was tested and the error was around 0.001 for a segment. Keywords-component; Temporal Packet Loss, Packt Loss, Networks Performance,Binary time series, Markov chains, Bernoulli modeL, MPLS.