Information Centric Networks for Parallel Processing in the Datacenter

Datacenters provide an interesting environment for the deployment of information-centric networking (ICN) protocols. The innovative features of ICNs, such as network-layer caching of named data, when combined with new opportunities given by naming semantics, such as the selective forwarding of data processing queries and the on-path processing of partial replies, introduce a large degree of freedom in protocol design. The advantages provided by these features can be accentuated by an appropriate choice of underlying datacenter topologies. In this paper, we discuss the design of a Map/Reduce application protocol built according to ICN naming semantics, consider its requirements in terms of support by the networking substrate, and briefly evaluate its behavior with respect to caching and aggregation in a highly-dimensional datacenter topology.