Information-Centric Networks for Distributed Computing

Information-centric Networking (ICN), a paradigm that operates on self-contained named data, naturally extends the role and capabilities of in-network elements. The flexible naming of ICN can be used to convey instructions for content routers to perform a target computation. In this paper, we consider ICN's potential of integrating distributed computing frameworks with the network layer in datacenter scenarios. By representing computation requests as named data, we aim to seamlessly integrate processing - executing a set of operations on a target input to generate the expected output - and storage - retrieving data and the cached output of previous processing requests - with the network, enabling computations to be distributed among nodes on the request's network path. Here we consider simple examples of distributed computation models that can be supported using an ICN syntax. We discuss mechanisms by which appropriate naming can oversee the distribution of a computation across network topologies, present the advantages of ICN integration, and survey the open research challenges.