Data Mart > MPP > Machine Deployment > How to Deploy a MPP Multi-machine Data Mart System

Suppose that a cluster of 4 nodes, and the machine is equipped with a 2-route 10-core CPU, and 32G memory, it can support real-time analysis of around 1TB data. Due to the different data features of various industries, the estimation factor should be adjusted accordingly.

For one distributed data mart system, only one machine is for Naming, multiple machines can do Map and Reduce, and multiple machines can be supported for Client. One machine can be a combination of multiple types. According to the number of machines,configuration,data traffic and other indicators, users can consider defining the type of node of machines.

➢For example: under the current circumstance, adopt one machine to play four roles, and the rest three machines only make Map node. For the machine doing four roles, a machine of better configuration can be selected. During the installation of Yonghong software, check MPP multi-machine version data mart, and tick four roles. Fill the required parameters according to instruction. Install Yonghong software on other three machines, check MPP multi-machine version data mart, and check the role of Map node.

Please also state that the Naming node can be cold back-up.

dc.fs.naming.paths=c\:/bihome/cloud/qry_naming.m;\\192.168.2.99/bihome/cloud/ qry_naming.m

Here use ";" to separate the metadata files under two paths, if the first machine is down and the disk crashes, the metadata information on the second machine can be manually obtained, and a new Naming node will be started.

After the installation, please confirm that access permission is given to the installation package. If it is under linux,it's best to start the program as a superuser.

Once all machine nodes are started, the data can be extracted and analyzed according to the documents.