Dexferizer: A service for data transfer optimization

Abstract

We present an approach to optimizing the transfer of data objects within a user's collection of computers and personal devices, subject to a variety of user-defined quality metrics such as cost, power consumption, and latency. By abstracting object transfer as a high-level service, and employing declarative networking techniques to cast object transfer as a constrained optimization problem, we show how to transparently exploit techniques as diverse as swarming and multi-hop transfer through virtual machines. Using a data replication system as a driving application, we demonstrate that our approach can easily accommodate flexible policies not easily implemented with existing solutions, and can thereby result in savings in time, power, bandwidth, or other costs.