Tight bounds for the performance of longest in system on DAGs

  • Authors:
  • Micah Adler;Adi Rosén

  • Affiliations:
  • Department of Computer Science, University of Massachusetts, Amherst, MA;Department of Computer Science, Technion, Haifa 32000, Israel

  • Venue:
  • Journal of Algorithms
  • Year:
  • 2005

Quantified Score

Hi-index 0.00

Visualization

Abstract

A growing amount of work has been invested in recent years in analyzing packet-switching networks under worst-case scenarios rather than under probabilistic assumption. Most of this work makes use of the model of "adversarial queuing theory" proposed by Borodin et al. [J. ACM 48 (1) (2001) 13-38], under which an adversary is allowed to inject into the network any sequence of packets as long as--roughly speaking--it does not overload the network.We show that the protocol Longest In System, when applied to directed acyclic graphs, uses buffers of only linear size (in the length of the longest path in the network). Furthermore, we show that any packet incurs only linear delay as well. These are, to the best of our knowledge, the first deterministic polynomial bounds on queue sizes and packet delays in the framework of adversarial queuing theory (other than on trees and the cycle). Furthermore these results separate Longest In System from other common universally stable protocols for which there exist exponential lower bounds that are obtained on DAGs. Our upper bounds are complemented by matching linear lower bounds on buffer sizes and packet delays.