Scale-out data processing systems can leverage high-speed networks to achieve performance gains. However, upgrading to faster networks, including making use of remote direct memory access (RDMA) – a low overhead communications protocol that provides low-level abstractions called RDMA verbs – is challenging because it often requires several decisions about low-level details). To help the developers of data processing systems exploit high-speed networks, Thostrup et al. propose DFI, which represents data movement as flows. DFI flows enable developers to declaratively specify how data should be routed for distributed data processing tasks and define optimization hints without engaging the low-level complexity of network communication.