An Analytics-Based Envelope Neural Net Approach to Forecasting Memory Leakage in an Enterprise Applications Environment

Abstract

This paper investigates a model to forecast memory-leak behavior in a big applications environment with short time frames, using neural networks. This process predicts the likelihood of a memory leakage in an IT infrastructure, with large number of enterprise applications running in a real world, while analyzing the work-load demand patterns hidden in the transactional data of the enterprise applications. This approach generates synthetic workload profiles and then based upon extracted insights, predict memory leakage behavior under peak load and stress conditions. The approach is investigated also for capacity tuning of the existing infrastructure and future planning of the required infrastructure. Co-learning, is used to train models and their use in analyzing the workload patterns. The workload prediction has enormous potential to optimize resources with minimal risk.