Exploiting Availability Prediction in Distributed Systems

Review

Problem Statement and Proposal

Distributed systems have significant scale and cost advantage over traditional architectures, but still the problem with availability stays varies widely. Availability modeling still stays crucial. Authors presented new techniques for predicting availability.

They applied this techniques by means of three application: availability-guided replica placement, improvement routing in delay-tolerant networks, combining availability prediction with virus modeling.

Related Works

Mostly all the related works are about empirical studies of availability in distributed systems. Lots of them are trying to detect uptime changes. The other technique is to use operating system logs to infer downtime. Also quite interesting research was done on the worse case scenarios, based only on hosts on-line time at night.

Although these scenarios cannot predict individual node behavior well.

One of the traditional approach for availability prediction was performed. The main idea was to fill data about uptime traces to well-known statistical distributions.

Schwarz proposed a distributed object store which biased data storage towards peers with high predicted availability. How is it done? Each node has a counter initiated on the value 0. During some specified time system scan is made and counter either increased or decreased for on-line and off-line system respectively.

Results

Authors introduced new techniques for availability prediction. To characterize differences in availability between DS, authors used techniques from signal processing and information theory. Authors performed great number of experiments on PlanetLab, Microsoft Nodes and Overnet. They proved that, by biasing replica storage towards highly available nodes, reduction in bandwidth consumption is possible. Also message latency could be reduced using authors uptime predictors.

decentralized_storage_systems/availprediction.txt · Last modified: 2012/04/23 01:07 by julia
 
Except where otherwise noted, content on this wiki is licensed under the following license: CC Attribution-Share Alike 3.0 Unported
Recent changes RSS feed Donate Powered by PHP Valid XHTML 1.0 Valid CSS Driven by DokuWiki