A generic availability model for clustered computing systems
Authors:
- H. SunHSH. Sun
- J.J. HanJHJ.J. Han
- H. LevendelHLH. Levendel
- 14 November 2002
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 241
Abstract
In this paper, we study the availability of clusteredcomputing system with one cluster manager and "N+M"processing nodes, where M processing nodes serve asspares for the N active processing nodes.The functionalityof individual processing node is dissected into applicationsoftware, management software, OS and hardware.Thedependency among these entities is considered.StochasticPetri net models are constructed to investigate the clusteravailability.In order to deal with the cluster with verylarge size, a solution based on state aggregation and fixed-pointiteration is proposed.The existence and uniquenessof the fixed point is proved.The impact of cluster manager,switchover time and coverage ratio are quantitativelystudied.From the numerical results of a simple clusterwith "2+1" processing nodes, we find that:(1) theavailability of cluster manager does not have significantimpact to the system availability, (2) system availabilityincreases with the coverage ratio and decreases with theswitchover time.The mechanisms to improve the systemavailability are discussed.Keywords
This publication has 8 references indexed in Scilit:
- An automatic fault diagnosis and correction system for telecommunications management2003
- Some research projects on clusters of personal computers2002
- NCAPS: application high availability in Unix computer clusters2002
- Reliability analysis of clustered computing systems2002
- Automated network fault management2002
- Software fault tolerance in a clustered architecture: techniques and reliability modeling1999
- Approximate availability analysis of VAXcluster systemsIEEE Transactions on Reliability, 1989
- Artificial Intelligence in Communications Networks Monitoring, Diagnosis and Operations1989