EVALUATING CONSISTENCY ON THE FLY USING YCSB
Résumé
Distributed data storage systems so-called NoSql systems are widely used when facing big data driven challenges. Due to the CAP theo- rem, most of these storage systems rely on Eventual Consistency and do not offer strong consistency by default. In order to offer better performance with acceptable level of consistency, these storage sys- tems offer different consistency options. These consistency options enable the user to choose different tradeoffs between Consistency, Latency and Availability. Hence selecting the right storage system for a given application requires performance evaluation and possible tradeoffs in advance.
In order to help future architects to choose the right system and corresponding tradeoffs for their application, this paper describes an extension of the YCSB: Yahoo cloud Services Benchmark that can report the number of stale reads in real time along with its default metrics. Evaluating the number of stale reads of a distributed data storage in real time system is a challenging task. Most of existing techniques found in the literature rely on log processing after the test is done, an approach that can be less practical instead of evaluating it at the time of test suite.