Traffic control mechanisms for data center networks

基于数据中心的云计算已经引起了工业企业、服务提供商和各级政府的广泛关注。使用商用组件构建的数据中心在实现可扩展性、最大限度地降低客户成本和节约能源消耗方面具有巨大潜力。然而，在充分实现新数据中心设计范式的优势之前，必须解决许多挑战。Microsoft最近对包含6000台服务器的生产集群进行的测量表明，由各种流量模式引起的拥塞可能会导致严重的损害，从而导致高应用程序延迟和大量吞吐量损失。为了帮助云计算服务提供商为其数据中心网络实现令人满意的性能，我们建议对数据中心流量控制算法进行系统的研究。该项目将开发基于不同路径的新机制和分布式算法来解决各种拥塞问题。将与现有解决方案进行比较，研究拟议机制的性能。将建立原型来测试所提出的解决方案的可行性。

Cloud computing based on data centers has attracted wide attentions from industrial companies, service providers, and various levels of governments. Data centers built with commodity components have great potential on achieving scalability, minimizing costs for customers, and saving energy consumption. However there are many challenges that must be resolved before the benefits of the new data center design paradigm can be fully realized. Recent measurements of a 6000-server production cluster by Microsoft have shown that congestions caused by various traffic patterns can cause severe impairments leading to high application latencies and significant amount of throughput loss. In order to help cloud computing service providers achieve a satisfactory performance for their data center networks, we propose to conduct a systematic investigation of the state-of-the-art traffic control algorithms for data centers. New mechanisms based on diverse paths and distributed algorithms will be developed to solve the various congestion problems with this proposed project. Performance of the proposed mechanisms will be studied in comparison with the existing solutions. Prototype will be built to test the feasibility of the proposed solutions.