Administrative Concepts

A set of cooperating processes that distribute data across a set of host computers.

Three kinds of cooperating processes implement a data grid: nodes, proxies, and state keepers.

A data grid partitions the complete set of data into copysets. Each copyset contains a portion of the full data set.

Each table row resides within only one copyset.

The data grid horizontally partitions the rows of a table across copysets. So a query or a transaction could span many copysets.

Nodes are processes that implement a copyset. Administrators define nodes and assign them to copysets.

Each copyset requires a primary node. Secondary nodes can provide optional backup protection.

Each node of a copyset maintains one copy of the data (that is, one copy of all the rows in that copyset).

Each node is part of only one copyset.

• In a prototyping or testing environment, you can implement a copyset using only one node.
• In most production environments, two nodes provide adequate protection.
• For even stronger fault tolerance, you can use three nodes.

The replication feature, when used, provides fault tolerance by preventing data loss when a node (or the machine running the node) fails and cannot be accessed.

All nodes in a given copyset are replicas of each other and they all have the exact same set of data.

There is a single primary replica in a copyset and the other nodes in that copyset are secondary replicas.

Every copyset in the data grid is organized to make sure the slice of data owned by that copyset is stored on as many replicas as desired.

Application programs connect to proxies, which in turn connect to nodes.

Proxy processes are independent of one another and do not require persistent state, so you can share the load of operations among multiple proxies.

Fault-tolerant state keeper processes determine and record the data grid's run time state information by which a data grid operates, and supply this information to the proxies and copyset nodes.

• Nodes in different copysets are created with the goal of scaling horizontally.

  As a result multiple copysets are created, each with a slice of the data.

• Nodes in the same copyset are created to provide multiple replicas for fault tolerance.

  These contain identical copies of the data.

• In a product environment a combination of the previously described use cases can be used.

  For example, you might choose to have two replicas per copyset and multiple copysets (say three) to scale horizontally.

  In this example, your environment would have a total of six nodes.