MULTI-DIMENSIONAL PHYSICAL HOST RELIABILITY PREDICTION FRAMEWORK

Abstract
Cloud data centers face considerable challenges due to unexpected host failures ending up in service disruption, SLA violations & mounted operational costs. Conventional reactive fault-dealing procedures are inefficient in large-scale environments needing proactive reliability management. This paper suggests on novel multi-dimensional physical Host Reliability Prediction Framework capable of integrating the components of Host Health Monitoring System (HHMS), Dynamic Reliability Scoring (DRS) and Predictive Reliability Model (PRM). Methodological innovations of Correlation-Aware Temporal Feature Engineering (CAFTE), Content-Aware Dynamic Reliability Scoring (CA-DRS) and Dynamic Ensemble Reliability Predictor (DERP). 92.3% prediction accuracy for 24-hour failure windows, 35% SLA violations reduction and 25% improved efficient reliability-aware VM placement was demonstrated by 500+ host testbed experiments. This framework has achieved up to 15% better performance than the currently available other methodologies Keywords - Data Center Management, VM Placement, Dynamic Ensemble Reliability Predictor, Machine Learning, Failure Prediction, Host Reliability Prediction, Cloud Computing