feature: Fine-grained control of the CPU.

[Super-long]

Signed-off-by: Super-long [email protected]

Describe what this PR does / why we need it

This PR attempts to improve the existing CPU burn implementation by modifying the coarse-grained loop to a finer one and introducing more algorithms to enrich the CPU burn.

Describe how you did it

The existing CPU fullload maintains the CPU load specified by the cpu-percent and gets q and ds based on the quota channel, then executes the CPU burn logic based on q and ds, I modified the implementation of the existing loop to introduce a richer and more diverse cpu burn algorithm, the algorithm ideas refer to the implementation of stress-ng --cpu-load.

Notes

The reason I greatly complicated burn's default judgment compared to previous implementations is that in extreme cases burn may get two consecutive values from quota, which can cause a sudden increase in CPU burn load. I encountered the following situation in the 910dab71b62ff91bbc715ccde4fc5faf59d274c1 commit:

TODO List

New features are still being implemented, I hope you have more comments.

1. More diverse CPU burn algorithms can be implemented in cpu_methods.
2. Do existing loop-based CPU burn implementations really need to be replaced? There is already a simple test for implementing CPU burn based on loop you can refer to the Specify the CPU percentage section in https://blog.csdn.net/weixin_43705457/article/details/124904793

Test

1. First execute taskset -c 0 . /chaos_os create cpu fullload --uid=d32d49a509d73856 --cpu-count=1 --cpu-percent=70 --cpu-index 0 Observe that the current CPU0 load is about 70%.
2. Then execute taskset -c 0 . /stress-ng --cpu 1 --cpu-load 40 --timeout 20h and observe that the CPU usage of the stress-ng process is 40% and the CPU usage of the chaos_os process is 30%.
3. Shut down the last stress-ng process, execute taskset -c 0 . /stress-ng --cpu 1 --cpu-load 80 --timeout 20h and observe that the CPU usage of the stress-ng process is 80% and chaos_os is nearly zero.
4. Shut down the last stress-ng process and observe that the current CPU0 load is about 70%.

CPU burn

The current individual CPU burn algorithm's average consumption time, The test code is available here: https://github.com/Super-long/tproxy-http_hijacking/blob/main/stress_cpu.go

• fft 180.146µs
• pi 1.596478ms
• fibonacci 193ns
• ackermann 4.687225ms
• bitops 337.691µs
• collatz 4.216µs
• crc16 108.159µs
• factorial 61.788µs