Chameleon@Edge Community Workshop Report

The widespread availability of inexpensive Internet of Things (IoT) and edge devices, that can be deployed at scale and used to observe, measure, and interact with our environment, are revolutionizing the opportunities available to science. At the same time building systems that use those devices dramatically changes the assumptions we make about infrastructure, system properties, and applications, and thus requires significant research to make good on the promise of this new technology. As a scientific instrument for systems research, the Chameleon testbed is evolving to make such research possible and has developed a preview set of new capabilities allowing users to allocate, reconfigure, and experiment with edge devices (CHI@Edge). To assess the alignment of the this set of capabilities with researchers’ needs (i.e., the product definition for the testbed) the Chameleon team organized a community workshop where experimenters showcased projects developed on CHI@Edge and discussed testbed requirements needed for the advancement of current and future research.

The primary insights gained through the workshop is the evolution of infrastructure scientific instruments from focus on infrastructure ownership to focus on implementation of infrastructure management and sharing, user services, and federation. In particular, edge testbeds are likely to evolve to support mixed ownership of hardware (i.e., one in which one of the user roles is going to be that of hardware provider) due to importance of infrastructure placement, its configuration (specifically, association with IoT devices interacting with the environment), as well as affordability of the edge hardware. This shift of focus requires new features such as restricted hardware sharing, and new operational models taking into account not only innovative hardware types, but also its deployment context (i.e., new trade-offs in security, accessibility, and connectivity), and less specialized resource providers. The broad range of context configurations will have to take into account the variability of demands from various communities/contexts and potentially provide a spectrum of reconfiguration methods rather than rely on a one-size-fits-all approach. The ability to connect such lightweight, community-specific, and often ad hoc testbeds to larger pools of static shared network and datacenter resources via consistent interfaces was judged to be of paramount importance, as were facilities for sharing research in an actionably consumable way (i.e., via digital forms such as data, software, or packaged experiments) so that different groups can leverage results produced by others more easily.