An RNN-LSTM Based Flavor Recommender Framework in Hybrid Cloud

摘要

One of the key problem in hybrid cloud is to discover well matched cloud provider to scale up applications irrespective of their non-standardized naming technologies. No framework has been developed to monitor and recommend VM flavor for the period of autoscale in hybrid cloud based on utilization history. In the existing scenario, administrators manually consider heterogeneous sets of criteria and resource relationships to map cloud service providers for user's preferences in hybrid environment. Flavor selection remains constant irrespective of application's resource usage in hybrid cloud which results in under utilization. The proposed framework will fill the gap by monitoring applications and recommending flavor to adjust capacity of resources at low possible cost while maintaining stability and predictable performance. The framework a) Predicts future workload using deep learning technique Recurrent Neural Network (RNN) with Long Short Term Memory (LSTM) b) Recommend a flavor, aligned with Optimized Cost and Service Level Agreement (SLA) for autoscale group depending on CPU or RAM utilization in the current and history workloads. c) Operate recommended flavor on future workload and ensures zero application downtime in the current workload. The proposed flavor recommender framework has been implemented in hybrid cloud OpenStack and Amazon Web Services (AWS). The experimental results has shown significant cost difference of 17.65% per hour on autoscale group of instances with proposed flavor recommender framework over traditional flavor selection.