Tutorial on Epsilon Serializabilityand a Survey of Recent

Epsilon Serializability (ESR) is a generalization of classic serializability (SR). ESR supports more concurrency and autonomy than SR by allowing a bounded amount of inconsistency in transactions that can tolerate it. ESR has three main advantages over previous \weak consistency" models: (1) ESR is a general framework, applicable to a wide range of application semantics; (2) ESR is upward-compatible, since it reduces to SR as ! 0; and (3) ESR has number of eecient algorithms that support it. Since its introduction 17], several papers have been published on ESR describing its properties, applications, algorithms, and performance evaluation. We brieey describe the concept of ESR, summarize the published work on ESR as well as ongoing research, and outline the future research directions. An important assumption in ESR is the existence of a distance function and an associated regular geometry in the database state space. Fortunately, many of the real world database state spaces fullll this requirement. For example, integers and real numbers in banking, airline, and scientiic data are cartesian spaces that have a natural deenition of distance function (the diierence) and the regular geometry of a metric space 11]. EEcient algorithms that support ESR use these properties to preserve the inconsistency bounds. The management of inconsistency in database state spaces with irregular geometries remains an interesting area of research. Another dimension in the ESR research is the algebraic properties of transaction operators. A classic transaction is deened as a sequence of operations that transform a consistent database state into another consistent state. This deenition bypasses the issue of what operators are employed within a transaction. In ESR work, a bounded amount of inconsistency may be introduced into a transaction and we are interested in how this inconsistency is propagated within the transaction as well as in the database. This line of investigation 21, 4] beneets from previous work on operation semantics (e.g., 2], cite UCSB work as a potential example). The results will help application designers in estimating the amount of inconsistency tolerated by the applications. Unlike the previous work on operation semantics, however, our algorithms do not depend on particular operation semantics. Therefore, we have database state space geometric properties and ESR algorithms on the system side, contrasting with transaction operator algebraic properties on the application side. The application interface to ESR systems is an extension of classic transactions, called Epsilon-Transaction (ET). Each ET may specify the bound on …