Managing Resources in Extensible Networks

Sumi Choi
Washington University

Thusday, February 19, 2:00PM
Lieb 3rd floor Conference Room
Computer Science Department
Stevens Institute of Technology
 

Abstract


Programmable networks allow easy configuration of resources for applications in a networked environment. Using programmable networks, an application can be designed to install and run its customized program module on remote network nodes such as routers. For example, programmable networks would enable easy design of a secure transmission application that allows data coming from an end device to be encrypted before it leaves the sender's domain and decrypted after it arrives in the receiver's domain. Resources required by a session in a programmable network include links and processing resources. In this talk, I will present solutions to two problems related to resource configuration in programmable networks.

First, we discuss the problem of finding the most cost efficient set of resources to configure a session. Particularly for unicast sessions which has two end points, this problem resembles usual routing problems that are often solved by the shortest path algorithms. In our setting, the processing requirement distinguishes it from those problems, and the goal becomes finding a path, or configuration, between a given end points that also includes processing nodes. We present an algorithm that finds the optimal (least cost) configuration given two end points and a processing requirement.

Second, we study how to provision a programmable network. In order to handle traffic from such applications, network resources including links and processing nodes should be dimensioned appropriately. We discuss the resource dimensioning problem in programmable networks by extending the one in conventional networks. We show how the technique used in conventional networks can be applied to programmable networks in a similar way. We also consider a more flexible situation where we have a freedom to place processing resources. We set our goal here to finding the set of processing locations that makes the most cost efficient resource dimensioning.