ODR: On-Line Delay Reduction Approach for Transferring Dynamically-Generated Big Flows Leveraging Software-Defined Networks

摘要

Data analytics techniques are changing the way businesses address core activities like planning, performance evaluation, and decision making. Recently, greater insights are reached as a result of applying such techniques to sets of data growing in diverse dimensions (big data). This attracted extensive research efforts to tackle big data transfer, processing, and storage challenges. Despite being similar in purpose to typical data transfer through the network, big data transfer poses significant changes. These changes warrant a new look at the techniques being used not limited to simple bandwidth increase. In this paper, we propose and implement an on-line algorithm to minimize the delay experienced by near real-time dynamically-generated big data flows when traveling through diverse networks leveraging Software Defined Networking (SDN). The algorithm exploits a comprehensive network view and statistics gathered by SDN controller to move big flows adaptively. The objective function is to minimize the network delay constrained to Service Level Agreement (SLA), in addition to network status. The proposed On-line Delay Reduction (ODR) approach outperforms the de facto Equal Cost Multi-Path (ECMP) algorithm with 19% to 24% reduction in delay.