Instability of sharing systems in the presence of retransmissions

From MaRDI portal
Publication:747714

DOI10.1007/S11134-015-9440-3zbMATH Open1325.60147arXiv1409.5622OpenAlexW2113025338MaRDI QIDQ747714FDOQ747714


Authors: Predrag R. Jelenković, Evangelia D. Skiani Edit this on Wikidata


Publication date: 19 October 2015

Published in: Queueing Systems (Search for Journal in Brave)

Abstract: Retransmissions represent a primary failure recovery mechanism on all layers of communication network architecture. Similarly, fair sharing, e.g. processor sharing (PS), is a widely accepted approach to resource allocation among multiple users. Recent work has shown that retransmissions in failure-prone, e.g. wireless ad hoc, networks can cause heavy tails and long delays. In this paper, we discover a new phenomenon showing that PS-based scheduling induces complete instability with zero throughput in the presence of retransmissions, regardless of how low the traffic load may be. This phenomenon occurs even when the job sizes are bounded/fragmented, e.g. deterministic. Our analytical results are further validated via simulation experiments. Moreover, our work demonstrates that scheduling one job at a time, such as first-come-first-serve, achieves stability and should be preferred in these systems.


Full work available at URL: https://arxiv.org/abs/1409.5622




Recommendations




Cites Work






This page was built for publication: Instability of sharing systems in the presence of retransmissions

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q747714)