I was following some discussion in tanglemath the other day and it was said, that full nodes are not needed and that swarm nodes will represent 99% of all nodes. Which brings me to the question: What exactly are swarm nodes ? Why are they better than full nodes ? Who sets these swarm nodes up ?
IOTA is the only partition-tolerant DLT. This means that IoT devices do not need to store the entire DB. Snapshotting is also used to keep the local database small. Swarm nodes will allow small IoT devices to run their own "micro" nodes which together with other IoT devices can form a swarm node.
These micro nodes would then communicate with each other to acquire information about the parts of the Tangle which they don't have in their own database. A disadvantage is, that they would have to trust each other as they cannot check if the information sent by the other micro nodes is correct.
In response to the comment in the original question, Swarm Nodes are mentioned in the IOTA Roadmap found here: https://blog.iota.org/iota-development-roadmap-74741f37ed01
Swarms will allow mini-nodes to exist by connecting to each other and seeing small partitions of the tangle, allowing them to participate in the tangle without seeing the entire picture that full nodes can see. Swarms will be in a state of tangle with each other, connecting to form a bigger picture of the tangle.