Research And Implementation Of Monitoring System For Container-based Cloudlet

Cloud computing can deploy and perform resource-intensive tasks with massive resources quickly. But there are some applications, which require not only a large number of computing resources, but also low latency. However, most of the devices are connected to datacenter via WAN. Latency of network makes these applications low-efficient. There is a new cloud-cloudlet. Cloudlet, composed of some computing nodes, is located at the edge of Internet. On the one hand, the mobile devices can connect to cloudlet via wireless network. In this way, users can get enough computing resource, with low latency. On the other hand, most of cloud service utilize virtual machine to meet the demands of resource reuse and user isolation. Container technology can also meet these needs. And compared with the use of virtual machine, containers have higher resource utilization, closer performance to the physical machine, and. faster starting speed. These advantages of container technology make it become the first choice to build cloud system.The monitoring system is essential for guaranteeing the normal operation of cloudlet and service optimization. Cloudlet monitoring system includes two levels. Cloudlet provides PaaS service for users, and each container loads and runs an application component. The monitoring system provides monitoring for running state of application components. In order to ensure stable operation and efficient allocation of resources in cloudlet system, monitoring system needs to provide monitoring for host machine to understand the use of resources. The main work of this thesis is to research and development a monitoring system for typical container-based cloudlet.First of all, this thesis briefly introduces and analyzes the architecture, data acquisition, data transmission and data storage of existing monitoring system, and points out the advantages and disadvantages of different methods.Second, combining the characteristics of container-based cloudlet, this thesis designs and implements a monitoring system for it. This monitoring system adopts tree type structure. It is divided into two parts, monitoring agent and monitoring server. Monitoring agent is a leaf node, which is mainly responsible for collecting container information, physical machine information and service information. Also, it takes care of anomaly detection, data aggregating and uploading. Monitoring server is a non-leaf node. It is mainly responsible for data aggregating, which come from lower monitoring agent or monitoring server. It also monitors the running status of the subordinate nodes. At the same time, the thesis designs the reliability and expansibility of the monitoring and control.Then, because the bandwidth of the monitoring system tends to be limited at very low levels, monitoring system only upload the summary data step by step to reduce the bandwidth occupation. In order to ensure that users can monitor the containers of interest in real time with low bandwidth, we design and implement a monitoring data request aggregation algorithm to meet the low bandwidth constraints by avoiding the transmission of duplication of data and reducing the number of message headers.We has deployed the monitoring system and carried out some tests to it. The result shows that the monitoring system is well positioned to meet the needs of monitoring container-based cloudlet with high reliability and low load, and the request aggregation algorithm reduces the bandwidth to transmit the monitoring data effectively.