A RM-Based Static Deployment System for Cluster

Abstract

In this paper, a Reliable-Multicast (RM) based static deployment system for cluster is designed which can be functionally divided into three parts: image capture, image transference, and node configuration. Image capture takes the snapshot of the operating system in the source node. Image transference distributes the captured image to target nodes. Node configuration mainly finishes the configuration of some identified information to target nodes.

Applying the technology of Image-Based Installation [1] and an idea of separating the operating system image from service software, this system solves the tedious problems involved when the large-scale cluster is divided into a number of functional pools; using the network transmit model of RM, this system reduces the cost of network bandwidth and significantly improves the efficiency in deployment; combining WOL [2] and PXE [3], the system provides the function of single control point and lightens the workload of the system administrator. So this system can solve the problems of current cluster deployment mechanisms, such as inefficiency, imperfection of automated node configuration, and only supporting the Linux-based operating system deployment, etc.

The deployment system consists of the deployment server, image servers, and target nodes. The deployment server is the core of the entire deployment system, which provides DHCP and TFTP service to ensure that the deployment control program can finish the deployment of target nodes after the cluster nodes start up from the NIC. Image servers mainly store the images and applications which can be accessed by the deployment server or target nodes by the NFS service. Target nodes are the nodes which need to be deployed in the cluster.

Before the deployment process, the image of the source node and the service environment of the deployment server and image server should be prepared. During the deployment, the target node communicates with the deployment server by using PXE and WOL, and then the deployment server deploys the corresponding image and application service to the target nodes through the RM method. Finally, the target nodes finish their configurations.

In the same experimental environment, some experiments respectively for this system, clonesys, and a deployment system identified by X have been built. As proved in experiments, this system can build a cluster more quickly and efficiently.