In addition to traditional concerns such as throughput and latency, freshness is becoming increasingly important. To stay fresh, applications stream status updates among their components, which can congest the network if the update frequency is too high. Tuning to the right frequency is not trivial, especially in the presence of other flows, when network sharing becomes much more involved. Also, sophisticated tuning logic inevitably complicates the design of the endhost devices. In this paper, we take an alternative approach. Instead of tuning the update frequency at the end-host, we let the endhost send out updates at its own pace and control the freshness within the network. This In-network Freshness Control (IFC) scheme allows the network operator to improve freshness while providing a fine-grained trade-off with throughput. IFC leverages in-network compute resources to filter out obsolete information during transmission of status updates, while queueing other drop-averse traffic separately to provide high throughput. We provide an analytic study of IFC and then implement IFC as Linux kernel modules. Our experiments show that IFC outperforms existing queueing disciplines by improving both throughput (by up to 40%) and freshness (by up to 50%). IFC can easily be combined with existing methods, e.g., BBR and DCTCP, and is effective even in partial deployments.
Download