Scarlet - Cluster Architecture

The Jira cluster has an Hub-and-Spoke architecture comprising a Terracotta central server and one or more Jira application instances representing the different nodes in the cluster.

Here is a picture showing the typical architecture:

The web clients.

Web clients access the Jira cluster by pointing to the load balancer IP address.

The load balancer.

The load balancer directs web clients requests to a given Jira application node.

The Jira nodes.

Each Jira node receive requests from the load balancer and process them by communicating with the database, the shared filesystem and the Terracotta central server.
Please note that there's no communication between nodes.

The Jira database.

The database is used by all Jira nodes as a storage backend.

The Jira Network filesystem or Storage Area Network.

The shared filesystem, implemented via an NFS or SAN service, is used for sharing Jira attachments among nodes.

The Terracotta central server.

The Terracotta central server coordinates the cluster and stores shared application data in its local file system.
Please note that while there's no communication between Jira nodes, all nodes communicate with the central server: so, the network link between the two parts should be as fast as possible.