Browse by author
Lookup NU author(s): Dr Paul EzhilchelvanORCiD
The paper addresses the problem of reliably multicasting messages within a cluster, with an advertised latency bound that would be met with a preset confidence level. The system developed here offers, prior to an invocation, probabilistic guarantees on reliability and latency bounds and, post invocation, strives to achieve the offer chosen by the application. Pre-invocation guarantees enable destinations to instantly build useful common knowledge about delivered multicasts with a correctness probability close to one.Design challenges arise due to communication delays being not entirely predictable and pre-invocation guarantees having to be made based on what is feasible in the imminent future. They are addressed by attributing unforeseen delay fluctuations to random and non-malicious perturbations termed as the noise. Cluster-based assessments find the system rarely failing to meet its obligation, and often more robust and faster than the guarantees it offers. In addition to having design novelty and delivering useful functionality, the system can be extended into high-throughput, low-cost ordering services essential for collaboration and state machine replication.
Author(s): Ezhilchelvan P
Publication type: Report
Publication status: Published
Series Title: School of Computing Science Technical Report Series
Year: 2011
Pages: 21
Print publication date: 01/07/2011
Source Publication Date: July 2011
Report Number: 1267
Institution: School of Computing Science, University of Newcastle upon Tyne
Place Published: Newcastle upon Tyne
