Hi  Yinghao, Here are some detail information provided by Yihao. 1. Implement the lastest version of CPU SLI feature (Sevice Level Index) for rich container;    this new version show great advantages in performance and accuracy.    The CPU SLI features isolate CPU runtime metrics for rich container users. It allows users    to use commands like `top` or read pseduo files like /proc/loadavg in a container, and get    the expected information, just as if they do in a traditional virtual machine.    CPU SLI features have evolved several versions. The latest version overcame left bugs, and    it achieved both great stability and accuracy.    For now it covers these metrics.    * CPU usage (usr%, nice%, sys%, idle%, steal%, iowait%, irq%, softirq%, guest%）    * 1/5/15 load accounting    * 1/5/15 running load accounting (meaning uninterruptible excluded)    * nr_running, nr_uninterruptible, nr_migration    * sched latency histogram 2. Design and implement cpu isolation solution for function computing running on Daishu, which    enable 2000 microvm running on one machine which acceptable overhead (< 1%).    Dynamic load tracking contributes most of the overhead, especially considering that enormous    2000 microvms expand a both wide and deep group scheduling tree. Yihao managed to minimalize    both cgroup hierarchy and dynamic load tracking, to limit the overhead of scheduling those    microvms to less than 1%. 3. The core develper in plugsched project, which aims to upgrade linux kernel scheduler subsystem online.    more details please visit project repo: https://gitee.com/anolis/plugsched 4. Backport cpuidle-haltpoll feature to ANCK and optimize performance for virtual machine.    Guest haltpoll is a relatively novel technique (at that time Yihao backports it). And he made it    enabled widely both on Aliyun ECS and in Alibaba Corp.    Also on Aliyun ECS, he oberserved and tackled a poorly performed locking problem, that qspinlock    was not enabled at all. This problem dates back long. At that time he solved this problem. The same    problem still persists on Amazon EC2. 5. Proficient in problem analysis, be skilled using gdb, crash, pyGDB, qemu ... https://lore.kernel.org/all/20220428103726.108597-1-wuyihao@linux.alibaba.c… -- csp 在 2022/5/18 上午11:38, 谢英豪 写道: