Multicast and customized deployment of large-scale operating systems

摘要

With the recent paradigm shift of cloud computing, deployment of operating systems (OSs) onto a large-scale computer network is becoming necessary. Note that there are usually numerous nodes with various functions in a cloud computing system. Thus, it is usually required to deploy different operating systems onto different nodes. In such a customized setting, conventional techniques of using unicast deployment to distribute a massive cloud OS onto thousands of nodes is time consuming and bandwidth-intensive. In this work, we thus propose a multicast deployment approach so as to significantly improve deployment efficiency. Furthermore, our multicast deployment approach can leverage existing configurations of the unicast counterpart. Specifically, the advantageous features of the proposed approach include the support of a reliable multicast protocol, a heterogeneous infrastructure, and cloud hypervisor environments. To evaluate the feasibility of the proposed approach in practical applications, CentOS and Ubuntu are used when implementing our deployment approach on several tens of nodes. Empirical studies show that both the required time for the entire distribution process, i.e., from starting delivery until the OS is ready, and the network bandwidth consumption are significantly reduced as compared with conventional unicast approaches. Consequently, less effort is required on monitoring and maintenance for system administrators.