An nameless reader shared this report from ZDNet:
On the Linux Plumbers Convention, the invite-only assembly for the highest Linux kernel builders, ByteDance Linux Kernel Engineer Cong Wang, proposed that we use AI and machine studying to tune the Linux kernel for the utmost outcomes for particular workloads… There are millions of parameters. Even for a Linux skilled, tuning them for optimum efficiency is a protracted, arduous job. And, after all, totally different workloads require totally different tunings for various units of Linux kernel parameters… What ByteDance is engaged on is a primary try to automate your entire Linux kernel parameter tuning course of with minimal engineering efforts.
Particularly, ByteDance is engaged on tuning Linux reminiscence administration. ByteDance has discovered that with machine studying algorithms, resembling Bayesian optimization, automated tuning might even beat most Linux kernel engineers. Why? Effectively, the concept, as Wang wryly put it, “is to not put Linux kernel engineers out of enterprise.” No, the purpose is “to liberate human engineers from tuning efficiency for every particular person workload. Whereas making higher choices with historic information, which people typically wrestle with. And, final, however by no means least, discover higher options than these we give you utilizing our present trial and error, heuristic strategies. Briefly, ByteDance’s system optimizes useful resource utilization by making real-time changes to issues like CPU frequency scaling and reminiscence administration.