On using circuit-switched networks for file transfers

  • Authors:
  • Malathi Veeraraghavan;Xiuduan Fang

  • Affiliations:
  • University of Virginia;University of Virginia

  • Venue:
  • On using circuit-switched networks for file transfers
  • Year:
  • 2008

Quantified Score

Hi-index 0.00

Visualization

Abstract

Ever since packet switching was invented in the seventies, it has been viewed as the better option for file transfers when compared to circuit switching. However, in recent years, owing to the lower costs of circuit switches at high speeds (10 Gb/s and higher), there has been an emergent interest in using circuit-switched networks for file transfers. While packet-switched networks offer proportional fairness giving all ongoing flows an equal share of link bandwidth, circuit-switched networks offer temporal fairness, i.e., giving deference to job seniority, enabling networks to offer users predictable service. In this dissertation, we present solutions to theoretical and practical problems related to supporting file transfers on dynamically shared circuit-switched networks. Given that file transfers do not have an intrinsic bandwidth requirement (the higher the rate, the lower the delay), a key question is what circuit rate should be allocated per file transfer. From our theoretical study, using analytical and simulation models, we recommend that at low loads, a homogeneous rate allocation should be made for all file transfers, while at high loads, a heterogeneous rate allocation, with higher-rate circuits for large files and lower-rate circuits for smaller files, yields an optimal combination of low mean response time while maintaining fairness between small-file and large-file calls. The practical problem we studied is how to integrate a dynamically shared circuit-switched network into the current connectionless packet-switched Internet. We designed a new gateway for this internetworking problem, one in which all sublayers of the network layer of the OSI protocol reference model are implemented. We characterize the performance of our gateway with tests on a local testbed as well as our high-speed, wide-area, experimental testbed called CHEETAH.