Analyzing Availability of Replicated Database Systems
Bharat Bhargava, Sumi Helal, Karl Friesen · Purdue e-Pubs (Purdue University System) · 1991
Qualitative studies have shown that replication control methods vary in the availability and performance of distributed database processing.Quantitative evaluation of these methods, however, requires a general availability model and experimental performance data.In this paper.we define and study aLgorithmic and operationaL availabilities of distributed database systems that employ replication as a technique to achieve fault-tolerance.We show that both availability definitions are complementary and therefore, should be studied simultaneously.We present a customer-stationary availability measure that includes four basic sets of parameters in its underlying model.These are transaction, data, configuration and failure parameters, in addition to parameters pertinent to the replication control method itself.We study algorithmic availability of the read-one-write-all and the quorum consensus replication control methods through a series of experiments in transaction processing of the RAID distributed database system.Operational availability of the same methods is studied using the SETH distributed database prototype and an event-driven simulator.