WebSphere eXtreme Scale Programming Guide > Access data in WebSphere eXtreme Scale > Data Grid API
DataGrid APIs and partitioning
With the DataGrid APIs, a client can send requests to one partition, a subset of partitions, or all the partitions in a grid. The client can specify a list of keys, and WebSphere eXtreme Scale determines the set of partitions that are hosting the keys. The request is then sent to all the partitions in the set in parallel and the client waits for the results. The client can also send requests without specifying keys, therefore, requests are sent to all partitions.
Agents that are deployed to the grid do not work in client mode. These agents work directly against the primary shard. Working directly against the primary shard results in maximum performance, allowing tens of thousands or more transactions per second because the agent works with the data at full memory speeds. Working directly with the primary shard also means that an agent can only see data that is within that shard. This provides some interesting opportunities that cannot be done on a client.
A typical eXtreme Scale client must be able determine the partition from the transaction, because the client needs to route the request. If an agent is directly attached to a shard, then no routing is needed. All requests go against that shard. Because the agent is directly attached to a shard, data in other maps in the shard can be accessed without worrying about common partitioning keys, and so on, because no routing occurs.
Parent topic
Data Grid API