On the detection and prevention of consistency anomalies in multi-tier and cloud platforms
Kamal Zellag · eScholarship@McGill (McGill) · 2013
Modern information systems, consisting of an application server tier and a database tier, offer several consistency guarantees for accessing data where strong consistency is traded for better performance or higher availability. However, it is often not clear how an application is affected when it runs under a low level of consistency. In fact, current application designers have basically no tools that would help them to get a feeling of which and how many inconsistencies actually occur during run-time of their particular application. In this thesis, we present new approaches to detect and quantify consistency anomalies for arbitrary multi-tier or cloud applications accessing various types of data stores in transactional or non-transactional contexts. We do not require any knowledge on the business logic of the studied application nor on its selected consistency guarantees. Our detection approaches can be off-line or on-line and for each detected anomaly, we identify exactly the requests and data items involved. Furthermore, we classify the detected anomalies into patterns showing the business methods involved as well as their occurrence frequency. Our approaches can help designers to either choose consistency guarantees where the anomalies do not occur or to change the application design to avoid the anomalies. Furthermore, we provide an option in which future anomalies are dynamically prevented should a certain threshold of anomalies occur. To test the effectiveness of our approaches, we have conducted a set of experiments analyzing the occurrence of anomalies in the benchmarks RUBiS and SPECj Enterprise 2010 under the multi-tier platform JavaEE and the benchmarks JMeter andYahoo! YCSB under the cloud platforms Google App Engine and Cassandra, respectively.