A novel model to improve network performance

Abstract

With the growing use of the Internet, as a wildly used tool, Internet has faced many resistances as an unreliable transmission medium. The core problem is time-varying delay in information transmission process. To solve this problem, we use the hidden Markov model (HMM) as a method of analysis first. We propose that the network model and hidden states are defined as the network’s situations. To train the HMM quickly, we use k-means clustering to obtain a more efficient starting value for determining hidden states, and we use Bayesian information criterion values to measure the model’s performance. Following this, an inner model is put forward based on the latent Dirichlet allocation. The inner model can be used to describe the essential associations among the network’s situations and delay transformation. Generally, we use Network Simulator (ns-2) to run the simulation, with changes made to the TCP based on the HMM and the inner model. By more accurately predicting the performance of network situation factors, such as link utilization, the time delay and network noise can be reduced.

Keywords:

• Hidden Markov model
• inner model
• time delay
• prediction