Global Mechanistic Computer Network Traffic Models with Applications to Network Cokriging

• Activity Number: 175
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Joel Vaughan*+ and Stilian Stoev and George Michailidis
• Companies: and University of Michigan and University of Michigan
• Keywords: Computer Networks ; Cokriging

Abstract

The packets of traffic sent across computer networks have a complex set of properties, such as the protocol used and the size of the packet. Different protocols induce different properties in the aggregate traffic load on the individual links that make up a computer network. For example, some protocols induce return traffic, while others do not. Because packets vary in size, the aggregate traffic traversing any link may be measured in terms of either the number of packets or number of bytes transmitted in a given time interval. We present a probabilistic model that incorporates these two important features: the protocols and the packet sizes. This model, although simple, incorporates both the dependence between forward and reverse traffic flows as well as the relationship between packet and byte measurements. We conclude with an application of the model to network cokriging, in which the packet and byte counts measured on an observed subset of the links in the network may be used to accurately predict the traffic levels on the other, unobserved links in the network. This application allows accurate monitoring of an entire network under resource constraints.