OmniData
It is a good practice that you use a storage-compute decoupled networking architecture to deploy OmniData. Such an architecture facilitates storage and compute collaboration, that is, operators can be pushed to storage nodes for nearby computing, improving big data computing performance. If OmniData and OmniOperator are used together, you are also advised to plan the network based on this section.
The environment planned in this document for operator pushdown consists of seven servers for example, which are one management node, three compute nodes, and three storage nodes. The decoupled storage-compute network is used. In the following, we will be using Ceph as the storage node for illustration:
- The management node is server1 for managing tasks.
- The compute nodes are agent1, agent2, and agent3, for running the big data query engine service.
- The storage nodes are ceph1, ceph2, and ceph3, for running the OmniData service.
OmniData executes operator pushdown (operator offload) and receives tasks pushed down by the Host Runtime. Therefore, the nodes on which OmniData is deployed are the offload nodes. In this document, the host architecture is the CPU architecture and OmniData is deployed on storage nodes, which are offload nodes actually.
A server can function as a management node, a compute node, and a storage node at the same time. In single-node mode, operations performed on the management node, compute node, or storage node (offload node) mentioned in the following sections are performed on the same node. Figure 1 shows the networking diagram.
