Mixture model approach for network forecasting

Abstract

Disclosed are various embodiments that provide a mixture model approach to network forecasting. Network traffic models are generated for multiple host types. Weights are determined for the network traffic models. A network forecast is generated based at least in part on a hardware footprint forecast and on the network traffic models as weighted by the determined weights.

Therefore, the following is claimed: 
     
         1 . A system, comprising:
 at least one computing device; and   a mixture model forecasting engine executable in the at least one computing device, wherein when executed the mixture model forecasting engine causes the at least one computing device to at least:
 generate respective network traffic models for individual ones of a plurality of host types; 
 determine respective weights for individual ones of the respective network traffic models; and 
 generate a network forecast based at least in part on a hardware footprint forecast and on the respective network traffic models weighted by the respective weights. 
   
     
     
         2 . The system of  claim 1 , wherein the hardware footprint forecast defines a predicted number of rack units in at least one data center. 
     
     
         3 . The system of  claim 1 , wherein the network forecast indicates a measure of network capacity to handle predicted network traffic between a first data center and a second data center that are covered by the hardware footprint forecast. 
     
     
         4 . The system of  claim 1 , wherein the plurality of host types include at least one of: a host type optimized for data storage, a host type optimized for computation, or a host type optimized for system memory. 
     
     
         5 . The system of  claim 1 , wherein when executed the mixture model forecasting engine further causes the at least one computing device to at least:
 determine respective Akaike information criterion (AIC) values for individual ones of the respective network traffic models; and   exclude at least one of the respective network traffic models from consideration in generating in the network forecast based at least in part on the respective AIC values.   
     
     
         6 . The system of  claim 1 , wherein when executed the mixture model forecasting engine further causes the at least one computing device to at least:
 determine respective Bayesian information criterion (BIC) values for individual ones of the respective network traffic models; and   exclude at least one of the respective network traffic models from consideration in generating in the network forecast based at least in part on the respective BIC values.   
     
     
         7 . The system of  claim 1 , wherein when executed the mixture model forecasting engine further causes the at least one computing device to at least:
 determine respective error variance values for individual ones of the respective network traffic models; and   wherein the respective weights are determined based at least in part on the respective error variance values.   
     
     
         8 . The system of  claim 1 , wherein when executed the mixture model forecasting engine further causes the at least one computing device to at least:
 determine respective overall measures of discrepancy using two disjoint training data sets for individual ones of the respective network traffic models; and   wherein the respective weights are determined based at least in part on the respective overall measures of discrepancy.   
     
     
         9 . The system of  claim 1 , wherein the respective weights are determined based at least in part on a plurality of randomizations of a training data set. 
     
     
         10 . A method, comprising:
 generating, by a computing device, respective network traffic models for individual ones of a plurality of host types;   determining, by the computing device, respective weights for individual ones of the network traffic models; and   generating, by the computing device, a network forecast based at least in part on a hardware footprint forecast and on the respective network traffic models weighted by the respective weights.   
     
     
         11 . The method of  claim 10 , wherein the network forecast is generated using a mixture model. 
     
     
         12 . The method of  claim 10 , wherein the network forecast comprises a time series specifying a respective predicted quantity for each of the plurality of host types. 
     
     
         13 . The method of  claim 10 , wherein the plurality of host types include a host type optimized for data storage, a host type optimized for computation, and a host type optimized for system memory. 
     
     
         14 . The method of  claim 10 , further comprising generating, by the computing device, the network forecast based at least in part on historical network traffic data associated with the plurality of host types. 
     
     
         15 . The method of  claim 14 , wherein the network forecast indicates predicted network traffic between a first network node and a second network node. 
     
     
         16 . The method of  claim 10 , wherein the hardware footprint forecast defines a predicted number of rack units in at least one data center. 
     
     
         17 . The method of  claim 10 , further comprising:
 determining, by the computing device, respective Akaike information criterion (AIC) values for individual ones of the respective network traffic models;   determining, by the computing device, respective Bayesian information criterion (BIC) values for individual ones of the respective network traffic models; and   excluding, by the computing device, at least one of the respective network traffic models from consideration in generating the network forecast based at least in part on at least one of: the respective AIC values or the respective BIC values.   
     
     
         18 . A non-transitory computer-readable medium embodying a program executable in a computing device, wherein when executed the program causes the computing device to at least:
 generate respective network traffic models for individual ones of a plurality of host types;   determine respective weights for individual ones of the respective network traffic models; and   generate a network forecast based at least in part on a hardware footprint forecast and on the respective network traffic models weighted by the respective weights.   
     
     
         19 . The non-transitory computer-readable medium of  claim 18 , wherein the respective network traffic models are generated using a first training data set, the respective weights are determined based at least in part on respective overall measures of discrepancy for the individual ones of the network traffic models, and the respective overall measures of discrepancy assess a respective accuracy of the individual ones of the respective network traffic models using a second training data set. 
     
     
         20 . The non-transitory computer-readable medium of  claim 18 , wherein when executed the program further causes the computing device to at least generate a forecast for networking hardware to handle the network traffic predicted by the network forecast.