MiCa: A Compositional Architecture for Gossip Protocols

The paper can be found here. and discusses a system to build management tools based on gossip protocols.

In general, gossip protocols are characterized by randomized peer selection and periodic communication, which makes them extremely fault tolerant since there is no realiance on a single path.

MiCa allows to compose several different gossip protocols and to run them independently

A protocols is defined by a tuple containing:

• Node’s states
• Rate of gossiping
• View: A probability mass distribution to sample the peers
• update of state

The framework supports different typs of composition

1. Independence preserving
2. Correlated: the gossip subprotocols are exectuted together as often as possible while presering the probability distributiond and the Rate

An example of composition is to first create a tree overlay, executed a broadcast opeation to the leaf nodes of the overlay and then aggregate the result back to the overlay root, i.e. a node acting as a leader.