Estimating and Forecasting Network Traffic Performance Based on Statistical Patterns Observed in SNMP Data

With scientific data growing to unprecedented volumes and the needs to share such massive amounts of data by increasing numbers of geographically distributed collaborators, the best possible network performance is required for efficient data access. Estimating the network traffic performance for a given time window with a probabilistic tolerance enables better data routing and transfers that is particularly important for large scientific data movements, which can be found in almost every scientific domain. In this paper, we develop a network performance estimation model based on statistical time series approach, to improve the efficiency of network resource utilization and data transfer scheduling and management over networks. Seasonal adjustment procedures are developed for identification of the cycling period and patterns, seasonal adjustment and diagnostics. Compared to the traditional time series models, we show a better forecast performance in our seasonal adjustment model with narrow confidence intervals.