Research In Key Technologies On Secure And Efficient Datacenter Networking

As a major direction for the next-generation computing, the cloud computing is based on large-scale data centers to acquire the capability of massive computing and data storage. Interconnecting tens of thousands of servers, data centers make use of advanced distributed computing technologies to provide cloud services as a warehouse-scale computer. Therefore, the performance of data center networks, which involves two key criterion, security and interactivity, determines the quality of hosting services.First, we propose a solution to the security-aware virtual network embedding problem. By analysing the potential security threats in data center networks, we summarize the security demands of virtual networks into three abstracted constraints. We combined these constraints with the basic virtual network embedding problem, and we formulate it as the security-aware virtual network embedding problem. Motivated by the node-ranking solution, we propose a heuristic algorithm to solve the problem in real-time. The efficiency and the effectiveness are validated through simulation.Second, we theoretically proved the effectiveness of RepFlow, an application layer solution to the long latency of some latency-sensitive flows in data center networks. Based on the fact that the service interactivity and thus user experience is greatly degraded by the long latency of some small flows, RepFlow replicates these flows to ensure a higher probability of quick delivery. By analysing the formulation of the queueing model, we propose RepSYN, an improved machanism, to avoid the performance degradation of RepFlow in extreme cases.Third, we implemented RepFlow and RepSYN as an API on the Apache Thrift and Node.js programming platforms, which make it possible for cloud service developers to use in an easy and compatible manner. Then, we test the flow completion time of latencysensitive flows on AWS EC2, a real data center experimental environment. The experimental results further prove the benefit of both RepFlow and RepSYN, and we conclude the choice conditions between them.