Scaling and Connection Distribution
Capacity is added by deploying more of the same components rather than by making one of them bigger, and user connections are distributed across them by the platform itself.
The agent does not choose its controller
An endpoint is not configured with a controller address. The sequence is:
- The agent connects to the platform's web service and the user authenticates - identity, geo check, device binding, device checks and the second factor.
- Only after that does the platform append the controller's domain name and port to the agent's configuration file.
- The agent then opens a DTLS tunnel to that controller.
See ISA Components.
Two things follow. The controller a user lands on is decided by the platform at sign-in, so distribution is a platform function rather than an endpoint setting or a client-side failover list. And a controller is only reachable by an endpoint that has already authenticated.
Controllers
More controllers are deployed as the user count grows - an active-active deployment, with each controller carrying part of the load. Deployments above a thousand users are sized this way, adding controllers in proportion to users rather than enlarging one.
A controller can also be held in active-passive standby: a second controller configured with the same network range and port, on separate infrastructure, ready to take over.
Gateways
A gateway has a published concurrent-connection capacity, which differs by gateway type - see Gateways. When the number of simultaneous users exceeds what one gateway supports, further gateways are provisioned and users are served across them.
Each gateway profile also names a Backup Gateway, so a gateway has a declared successor if it becomes unavailable - see Create Gateways.
Because applications are attached to gateways, where an application is published also determines which gateway carries its traffic - see Add Application to Gateway and What Traffic Goes Through the Tunnel.
Sizing
Capacity planning is therefore a count of components: controllers in proportion to concurrent users, gateways in proportion to simultaneous connections per gateway type, with a backup declared for each. The published per-gateway figures come from internal simulation and vary with the traffic a deployment actually carries, so they are a starting point for sizing rather than a guarantee.