Flow Scheduling Optimization For Data Center Networks

In recent years, the applications have begun to change our life and become the age of the Internet, such as cloud computing, big data, social networking. These applications all need data center support, so building high efficiency and low cost data center has become a hot topic in academia and industry. A growing number of users make the increasing scale of data center network, and there many new data center network topology are produced to enlarge the network bandwidth and throughput as to accept more users access. However, many applications make a new challenge to data center network.The applications produce much sudden and uneven distributed network traffic in data center network, which easily cause network congestion and degrade the application performance and influence the effect of user experience. How to guarantee the network flow evenly distributed in data center and non-blocking transmission between servers becomes the key to ensure the application performance. Our works is optimizing the network flow scheduling, which goal is improving the network availability and support large user scale. Our works mainly include three aspects:First, the data center server could improve the resource utilization for application based on the deployment of Virtual Machine (VM). The network flow between the VM performed communications between the servers in data center. In order to degrade communication between VMs on the network performance, the intuitive method is place VMs with the traffic large as much closer in the location. However, the emergence of various new cloud computing and social networking applications make communication flow between the VM is irregular and unpredictable. Unfortunately, this communication is inevitable when the VM is deployed. Some communication between the VMs may occupy more network bandwidth, which affects the other communication between the VM. How to avoid the bandwidth waste as the unreasonable position becomes important way to improve the performance of network. So we develop an improved dynamic bloom filter to detect data traffic between VMs, and define the communication cost function of the VM migration to evaluate VM migration. And then, a flow-scheduling algorithm based on VM migration is proposed to optimize the suitable VM migration and save throughput for the whole network throughput.Second, the modular data center expand the scale throughput the redundant paths between the servers to increase network bandwidth for file transferring and data distribution. However, the cloud and social networking applications produce uneven length and bursty flows, which can cause path congestion affects the efficiency of multi-path transmission. When the network utilization reaches higher level, the flow transmission cannot easily find a suitable path. To resolve this problem, we propose a adaptive scheduling strategy to calculate the suitable paths in modular data center network. Our method calculates more backup paths based on the end-to-end delay and the length of path. The greedy algorithm chooses the appropriate path. Out algorithm make progression in the low efficiency of multi-path transmission. When a network flow is coming, we choose the path with minimum overhead instead of the breadth-first algorithm. The evaluation and result show our method well support the requirement of the mixing flow transmission.Finally, the online social network application produces sudden and critical deadline flow. The flow characteristic and the requirement led the current flow scheduling method failure. Different with the traditional flow distribution in data center network, social network produce flow with different real-time demand, while users send messages and share pictures and videos. At the same time, the process of files storing on the server and backup produce much more long flows with high throughput requirement and non-real-time demand. The complex flow distribution make difficult for flow scheduling in current data center and easily degrade the network performance and user experience. So we propose a flow scheduling strategy based on priority to resolve this problem. We define the priority of flow based on the flow length and the deadline, and then, the appropriate transmission rate can be calculated based on priority. Our method meets the real-time requirement for mice flow and the maximum throughput of the network can be guaranteed at the same time.In summary, our researches propose a series of strategies to solve the problem of various performances from different angles for data center network. And the flow scheduling can effectively guarantee the performance of cloud computing and social networking applications.