Data Access Pattern Analysis and Prediction for Object-Oriented Applications
Stoyan Garbatov, João Cachopo · Americanae (AECID Library) · 2011
This work presents an innovative system for analysing and predicting the runtime behaviour of object-oriented applications, with respect to the data access patterns performed over their domain objects. The analysis and predictions are performed using three alternative stochastic model implementations. The models are based on Bayesian Inference, Importance Analysis, and Markov Chains. The system deals with all the necessary modifications of the target applications under analysis in a completely automatic fashion, without it being necessary for any developer intervention. The results are validated by the execution of the TPC-W and oo7 benchmarks. The oo7 benchmark has been modelled as a stochastic process through Monte Carlo simulations. We show that the results obtained with our system are precise, regarding the observed behaviour, and that the overheads introduced by the data acquisition are low, ranging from 5% to 9%. The system is sufficiently flexible to be applied to a broad spectrum of object-oriented applications.