Page History

...

Any peer member (either a application hosting a peer cache or a CacheServer) announces itself to the distributed system through one of two mechanisms:

...

by first contacting a Locator.  This is a component that maintains a discovery set of all peer members in the distributed system at any given moment. Though typically started as a separate process (with redundancy), locators can also be embedded in any peer cache (like a CacheServer). New members contact the locator via a TCP port to get the current set of peer members. These members are then asked who the current membership coordinator is. Locators are neither a bottleneck to cache operations nor a single point of failure.

...

Multicast based discovery: As an alternative to locators, members can use a multicast channel to find the membership coordinator and join the distributed system. Multicast is not enabled in all networks, particularly small, home-based networks.

The membership coordinator is usually the oldest member locator in the distributed system and is responsible for ensuring that all members know about changes in membership. If the membership coordinator fails, the GMS automatically selects the next-oldest member locator to function as the coordinator.  If all locators are down a server will become the coordinator but new members will not be able to join until at least one locator has been restarted.

Joining

Once a new process has discovered the membership coordinator's identity, it sends the coordinator a join-request. The coordinator responds by adding the process's identity to the membership set (called a membership view) and sending the updated set to the new process and to all of the existing members.

...