Within the overall structure of the cluster, the clients represent the application end-points. They work independently but can communicate through the active servers of the stripes to which they are connected.

Note that a client only ever interacts with an active server, never directly communicating with a passive server.

In a single-stripe cluster, each client is connected to the active server of that stripe. In a multi-stripe cluster, each client is connected to the active server of each stripe, interacting with them quasi-independently.

Within the logical structure of the cluster, the client isn't the process making the connection, but the connection itself. This means that a single JVM opening multiple connections to the same stripe will be seen by the stripe as multiple, independent clients.

When establishing a connection to a stripe, the client must find the active server. It does this by attempting to connect to each server in the stripe, knowing that only the active server will not reject the connection attempt.

If the active server to which a client is attached goes offline, the client will attempt to reconnect to one of the servers in the stripe. This is similar to what happens during its initial connection.

Note that there is no default time-out on this reconnection attempt. This means that it is possible for all clients to wait, blocking their progress, until a server is restarted, potentially days later.