Dear List, This is Qingyao Sun, an incoming CS Ph.D. student at UT Austin (unfortunately I specialize in machine learning instead of OS) who is brainstorming some project ideas for GSoC 2023. It looks like NetBSD has had Lua in its kernel since ~2010. I love the idea of a scriptable kernel, and working on Lua support in the NetBSD kernel sounds super exciting to me. I need a concrete plan for GSoC, so I would like to hear your opinion on which improvements can make Lua-in-kernel more useful and/or popular.
One idea of mine (inspired by LAKE [1]) is to showcase Lua's potential by providing an adaptative mechanism to fine-tune heuristic-based kernel subsystems, e.g. use an in-kernel Lua function to analyze the filesystem access pattern online and figure out the optimal amount of data to prefetch from disk, or even attempt to recognize some common non-sequential access patterns and prefetch accordingly. The LAKE paper employed machine learning and even a GPU (!) for the tuning, but I hypothesize that a ~100-line function in Lua on a CPU should already be sufficient for a significant improvement. The rationale behind this proposal is that Lua is more expressive and safe than C, so we can afford to make the business logic more complex when implementing the heuristics in Lua. Also, please let me know if anyone is interested in being a mentor for this project. Thanks! Bests, Qingyao ------ [1]: https://utns.cs.utexas.edu/papers/lake_camera_ready.pdf
