Buying Custom Clusters

Buying a Custom MRS Cluster

For details about customizing software and hardware parameters and advanced configurations of an MRS cluster, see Buying a Custom Cluster.

For details about how to store data in OBS and use an MRS cluster only for data computing, see Configuring a Storage-Compute Decoupled Cluster (Agency).

For details about how to adjust the deployment location of each component instance in the cluster to meet service requirements and prevent resource contention, see Buying a Custom Topology Cluster.

Cluster Types

• Analysis cluster

  Analysis clusters are used for offline data analysis and provide Hadoop components.

• Streaming cluster

  Streaming clusters are used for streaming tasks and provide stream processing components.

• Hybrid cluster

  Hybrid clusters are used for both offline data analysis and streaming processing and provide Hadoop components and streaming processing components.

• Custom cluster

  You can customize the deployment mode of clusters based on service requirements to ensure data management and control.

Node Types

• Master node

  Management node in a cluster. Master processes of a distributed system, MRS Manager, and databases are deployed on master nodes. Master nodes cannot be scaled out. The processing capability of master nodes determines the upper limit of the management capability of the entire cluster. MRS supports scale-up of master node specifications to provide support for management of a larger cluster.

• Core node

  An instance that processes and stores data. Core nodes are used for both storage and computing and can be scaled in or out. Since core nodes bear data storage, there are many restrictions on scale-in to prevent data loss and auto scaling cannot be performed.

• Task node

  An instance that process data but do not store cluster data (such as HDFS data). Task nodes are used for computing and do not store persistent data. It can be scaled in or out. Task nodes bear only computing tasks. Therefore, auto scaling can be performed. Yarn and Storm components are installed on task nodes. Task nodes are optional and the number of task nodes can be zero. When the data volume change is small in a cluster but the cluster's service processing capabilities need to be remarkably and temporarily improved, add task nodes. For example, you can add task nodes if long-term tasks must be completed in a short time.