Towards efficient load distribution in big data cloud

Abstract

Traffic spikes in big data cloud have led to great challenges for middlebox management in the behind datacenter networks, especially for load distribution among varieties of middleboxes. Contemporary flat load distributing strategies are inadequate to handle traffic spikes in terms of time and resource efficiency. Based on the spike patterns, a hybrid load distributing solution is proposed with hierarchical architecture, which conducts static distributing and dynamic collaborative distributing in respect of flow types. The evaluation results from the prototype system and simulation show that our solution is superior to existing load distributing strategies in terms of both time and resource efficiency.