On 3/19/26 03:13, Roy Fru wrote: > Dear Mentors, > > Please find attached my final report on the Linux Kernel Scheduler > Tracing Experiment. It covers kernel tracing, eBPF instrumentation, > and analysis of context-switch behavior under virtual CPU contention. > > While waiting for your feedback, I will continue analyzing > observations and exploring proposed directions for further study. > > Best regards, > Nchang Roy > Hi Nchang,
I enjoyed reading your report. The tasks listed under "Scheduler" are particularly challenging, so I appreciate the effort you have put into them. The tracepoint corresponding to scheduler tick events is a trick question on purpose. Good job ruling out the sched_stat_* tracepoints and considering a kprobe instead. The Experiment-0 task does not ask for a global average, so reporting per-CPU averages is actually a clever approach. However, please keep in mind that Phantom Tracker needs to report global averages, computed exactly once, by exactly one core, per scheduler tick. Please keep this requirement in mind as you refine your design. I am quite satisfied with your progress on Experiment-0, and I would like to invite you to submit a GSoC proposal for this project. Please refer to the thesis to understand /what/needs to be designed. In the proposals, I am primarily looking for the student's reasoning about /how/they plan to design it. Students are not expected to come up with the correct design entirely on their own. Rather, they are expected to discuss ideas, seek feedback, and ask clarifying questions. Since there are multiple competing proposals for this project, it is perfectly fine to share your drafts only with the mentors. I will be happy to answer any quick questions on the project IRC. I also plan to hold an office-hours-style call next week for students interested in contributing to this project. The other mentors may also join, subject to their availability. Himadri
