Network Traffic in a Cluster

There are some network considerations in a Swarm environment. Swarm nodes talk to each other over multicast on a particular, configurable multicast group (by default 224.0.10.100). It is our recommendation to keep all Swarm nodes on a separate VLAN so that inter-node traffic remains local to the VLAN. There is no reason to have other non-Swarm network-connected devices on the same VLAN as the Swarm nodes if they are not reading and/or writing to the storage cluster. Generically speaking, the nodes attach to a switch or group of switches that attach to a router. That router will provide routing access to the cluster. A router by default will block all multicast traffic so all of the inter-node Swarm cluster traffic will remain local to the VLAN. Certainly the nodes will need to be read from and written to by servers or client computers. Reading and writing data to a Swarm cluster from a server or client computer is done through standards-based HTTP commands over port 80 by default. Application servers that read and write to the Swarm cluster can have an interface in the same VLAN as the Swarm cluster for increased performance.

To be clear, multicast traffic between the nodes stays on the local VLAN and does not affect other VLANs. The router routes read/ write requests to and from the cluster via HTTP. Any application servers that have no need to contact the Swarm cluster should reside in their own VLAN.