Research On Traffic Management And Optimization In Data Center Networks

In recent years, cloud computing has attracted widespread concern from both industry and academia. At the same time, the number of new Internet online services users (such as web search, social networking, instant messaging, etc.) is growing rapidly. With the development of these services and use of new technologies, significant changes and trends are taking place in data centers and then bring new challenges and problems to data center networks. First, emerging services need a lot of one-to-many and many-to-many communication between the servers. The result is the rapid growth of internal traffic within data center which shows different features from Internet traffic. Under the current technical conditions, congestions frequently happen in data center networks, resulting in high packet loss rate, large latency and lower throughput, which affect application performance and quality of service. Second, virtualization technology has also brought new challenges to the data center network. Virtual machine migration traffic would increase the network load and even cause network congestion. Therefore, in order to ensure the performance of applications and the quality of services, traffic management and optimization in data centers has become an important issue to be solved.This paper first summarizes the research background and the latest progress of data center networking. Then, based on the characteristics of data center traffic, this paper studies traffic management and optimization in data centers, including traffic engineering, optimization and improvement of transport layer protocol and multiple virtual machine migration scheduling. Our objective is to reduce network congestion and optimize the utilization of network resources and enhance the performance of the business. The details as follows:First, traffic optimization of Ethernet link aggregation is considered. Ethernet link aggregation is a common scenario of data center networks but is ingnored by current current traffic engineering methods. The aggregated physical links are viewed as a logical link and the traffic distribution to physical links is not taken into account. Current session-based Ethernet link aggregation traffic allocation algorithm does not work well owing to the traffic characteristics of data centers. To address this issue, This paper first investigate the causes of frame disordering in link aggregation and find that all of them either are no longer true or can be prevented in data centers. Then we present a byte-counter frame-level traffic splitting algorithm which achieves optimal performance while causes no frame disordering. The only requirement is that frames in a session are the same size which can be easily met in data centers.Second, two prolems of TCP protocol in the data center networks, TCP scaling limit and TCP Incast throughput collapse, are analyzed. TCP which was originally designed for wide area network is not suitable for data center networks. This paper proposes to optimize the TCP performance by reducing the packet size. The smaller the packet size, link and switch buffer can carry more packets; then TCP congestion control is more accuracy and the packet loss is less. This can achieve the purpose of the improvement of TCP performance, including mitigating the TCP scaling limit and easing the the TCP Incast throughput collapse. Reducing packet size can be deployed easily in existing hardware and software and only simple configuration are needed on the servers or switcher.Third, a new protocol named DATCP (Deadline-Aware TCP) is proposed to provide deadline-aware transmission service for the commoditized data centers. Many flows in data centers have deadlines and missing deadlines would hurt application performance such as affecting response quality in web applications or delaying computing jobs in MapReduce-like systems. However, TCP which is widely used in data centers now cannot provide deadline-aware transmission service. Service differentiation only distinguishes flows with different priority but is unable to guarantee completion time. In this paper, we propose a DATCP to to provide deadline-aware transmission service for the commoditized data centers. DATCP dynamically adjusts congestion control parameters according to network condition. Simulation results show that DATCP can make flows meet deadlines effectively.Finally, VM shuffle scheduling in virtualized data centers is studied to minimize the impact to data center. Virtual machine live migration provides spatial flexibility by rearranging VM placement in several scenarios. However, VM live migration would consume scarce bandwidth and even cause network congestion. Since the bandwidth used by VM migration is usually the same as the services running in the VM, migration traffic would dominate network path and affect other application traffic as the traffic of a VM migration is usually several GBs. It gets worse in VM shuffle where plenty of VMs are needed to be moved. In this paper, we explore the opportunity to manage online VM shuffle and minimize the impact to data center networks. Two VM shuffle scheduling algorithms are presented to minimize the VM shuffle duration by coordinating VM migration in a proper scheduling order. VMs benefiting others maximally are migrated preferentially. Our evaluation shows that algorithms proposed in this paper decreases the shuffle duration dramatically.