Inter-Disciplinary Research Challenges in Computer Systems for the 2020s

The broad landscape of new technologies currently being explored makes the current times very exciting for computer systems research. The community is actively researching an extensive set of topics, ranging from the small (e.g., energy-independent embedded devices) to the large (e.g., brain-scale deep learning), simultaneously addressing technology discontinuities (End of Moore's Law and EnergyWall), new challenges in security and privacy, and the rise of artificial intelligence (AI). While industry is applying some of these technologies, its efforts are necessarily focused on only a few areas, and on relatively short-term horizons. This offers academic researchers the opportunity to attack the problems with a broader and longer-term view. Further, in recent times, the computer systems community has started to pay increasing attention to non-performance measures, such as security, complexity, and power. To make progress in this multi-objective world, the composition of research teams needs to change. Teams have to become inter-disciplinary, enabling the flow of ideas across computing fields. While many research directions are interesting, this report outlines a few high-priority areas where inter-disciplinary research is likely to have a high payoff: a) Developing the components for a usable planet-scale Internet of Things (IoT), with provably energy-efficient devices. This report envisions a highly-available, geographically distributed, heterogeneous large-scale IoT system with the same efficiency, maintainability, and usability as today's data centers. This planet-scale IoT will be populated by many computationally-sophisticated IoT devices that are ultra-low power and operate energy-independently. b) Rethinking the hardware-software security contract in the age of AI. In light of the recent security vulnerabilities, this report argues for building hardware abstractions that communicate security guarantees, and for allowing software to communicate its security and privacy requirements to the hardware. Further, security and privacy mechanisms should be integrated into the disruptive emerging technologies that support AI. c) Making AI a truly dependable technology that is usable by all the citizens in all settings. As AI frameworks automate an increasing number of critical operations, this report argues for end-to-end dependable AI, where both the hardware and the software are understood and verified. Further, AI needs to turn from a centralized tool into a capability easily usable by all the citizens in all settings to meet an ever expanding range of needs. d) Developing solutions to tackle extreme complexity, possibly based on formal methods. This report argues for the need to tame the explosion of system complexity and heterogeneity by creating new abstractions and complexity-management solutions. Such solutions need to be accessible to domain experts. An important step towards this goal is to scale out and extend formal methods for the real world. This report also describes other, related research challenges.