Biologically-inspired network 'memory' for smarter networking

Emerging technologies such as the Internet of Things generate huge amounts of network traffic and data which lead to significant challenges in a) ensuring availability of resources on-demand, b) recognizing emergent and abnormal behavior, and c) making effective decisions for efficient network operations. Network traffic data exhibit spatiotemporal patterns. Learning and maintaining the currently elusive rich semantics based on analyzing such patterns would help in mitigating those challenges. In this paper, we propose the concept of a network 'memory' (or NetMem) to support smarter data-driven network operations as a foundational component of next generation networks. NetMem will enable networking objects to understand autonomously, at real-time, on-demand, and at low cost semantics with different levels of granularity and related to various network elements. Guided by the fact that human activities exhibit spatiotemporal data patterns; and the human memory extracts and maintains semantics to enable accordingly learning and predicting new things, we design NetMem to mimic functionalities of that memory. NetMem provides capabilities for semantics management through uniquely integrating data virtualization for homogenizing massive data originating from heterogeneous sources, cloud-like scalable storage, associative rule learning to recognize data patterns, and hidden Markov models for reasoning and extracting semantics clarifying normal/abnormal behavior. NetMem provides associative access to data patterns and relevant derived semantics to enable enhancements in early anomaly detection, more accurate behavior prediction and satisfying QoS requirements with better utilization of resources. We evaluate NetMem using simulation. Preliminary results demonstrate the positive impact of NetMem on various network management operations.