Forecasting link utilization between points of presence in an IP network

A method for forecasting future link utilization demand between a pair of points of presence in an IP network, the method comprising:

• collecting prior link utilization information, the prior link utilization information identifying the aggregate link utilization between the pair of points of presence as a function of time;
• modeling the prior link utilization information using wavelet multiresolution analysis to create an approximation curve that models the long-term trend of the prior link utilization information;
• constructing at least one linear time series model of the approximation curve;
• evaluating each of the at least one linear time series models of the approximation curve to determine which model best matches the long-term trend of the prior link utilization information;
• and
• forecasting future link utilization using the linear time series model of the approximation curve that best matches the long-term trend of the prior link utilization information.

A non-transitory computer readable medium for causing a computer to perform a method for forecasting future link utilization demand between a pair of Points of Presence in an IP network, the method comprising:

• collecting prior link utilization information, the prior link utilization information identifying the aggregate link utilization between the pair of points of presence as a function of time;
• modeling the prior link utilization information using wavelet multiresolution analysis to create an approximation curve that models the long-term trend of the prior link utilization information;
• constructing at least one linear time series model of the approximation curve;
• evaluating each of the at least one linear time series models of the approximation curve to determine which best matches the long-term trend of the prior link utilization information;
• and
• forecasting future link utilization using the linear time series model of the approximation curve that best matches the long-term trend of the prior link utilization information.

A non-transitory computer readable medium for causing a computer to perform a method for forecasting future link utilization demand between a pair of Points of Presence in an IP network based upon prior link utilization information, the method comprising:

• modeling the prior link utilization using wavelet multiresolution analysis to create a plurality of approximation curves that combine to synthesize the prior link utilization information;
• using the analysis of variance technique to identify from the plurality of approximation curves:
• a long-term trend approximation curve that best matches the long-term trend of the prior link utilization information;
• a deviation approximation curve that models the bulk of the variability around the long-term trend of the prior link utilization information;
• constructing at least one linear time series model of the long-term approximation curve;
• selecting one of the at least one linear time series model of the long-term trend approximation curve that best matches the long-term trend of the prior link utilization information;
• selecting one of the at least one linear time series model of the deviation approximation curve that best matches the deviation around the long-term trend of the prior link utilization information;
• forecasting future link utilization using the selected linear model of the long-term trend approximation curve;
• and
• forecasting future deviation of link utilization demand using the selected linear model of the deviation approximation curve.