A practicable machine learning solution for security-cognizant data placement on cloud platforms

Abstract

While designing data placement strategies for cloud storage platforms, data security and data retrieval time are two equally important parameters that determine the quality of data placement. As these two parameters are generally mutually conflicting, it is imperative that we need to strike a balance between data security and retrieval time to assure the quality-of-service promised by the network/cloud service provider. To guarantee the data integrity of data stored on the network storage nodes in case of any threats or cyberattacks, the placement strategy should be adaptable to incorporate the threat characteristics. This is achieved by integrating machine intelligence to the network prone to attacks to identify the most vulnerable threat type for each node. This objective forms an imperative addendum to the attack resilient and retrieval time trade-off strategy (ARRT) strategy proposed in the literature to deploy as a practicable solution for a service provider. A set of Pareto-optimal solutions which strikes a balance between retrieval time and security based on inherent network properties by ARRT will be our initial condition for our machine learning model in this work. We take a radically different approach in which we attempt to identify the most vulnerable threat type for each node in the recommended Pareto-optimal solutions to minimize data loss through appropriate refinement of the existing data placement. This is achieved by supplementing the evolutionary algorithm with amachine learning model and we refer to this integrated and complete approach as security-cognizant data placement (SDP) strategy. In this study, based on the relevant performance metric that includes data integrity which is a measure of robustness, we evaluate and quantify our performance through rigorous discrete event simulations on arbitrary cloud topologies and demonstrate the impact of a neural network in delivering a superior performance.