search

gnuton / asuswrt-merlin.ng

Extends the support of Merlin firmware to more ASUS routers
Other
1.46k stars 83 forks source link

RT-AX92U reports high load average regardless of system load #530

Open pixdrift opened 9 months ago

pixdrift commented 9 months ago

Router Model Affected Models: RT-AX92U

Firmware Version Affected Linux RT-AX92U-0480 4.1.51 #2 SMP PREEMPT Fri Dec 29 23:56:03 UTC 2023 aarch64 ASUSWRT-Merlin

Is this bug present in upstream Merlin releases too? Unknown, will try and find references to similar on other systems.

Describe the bug Regardless of load on the system, the system load metrics consistently report close to full resource utilisation. Running on an RT-AX92U with 2 cores, the load average remains consistently at ~1.9 when the system is completely idle.

Mem: 246856K used, 161996K free, 4864K shrd, 0K buff, 27380K cached
CPU0:  0.2% usr  0.8% sys  0.0% nic 98.5% idle  0.0% io  0.0% irq  0.4% sirq
CPU1:  0.2% usr  1.0% sys  0.0% nic 98.7% idle  0.0% io  0.0% irq  0.0% sirq
Load average: 1.84 1.90 1.95 1/149 22582

There appears to be no discernible contributor to these load numbers, cpu usage, io wait, interrupts etc.

Note: This node is currently in an AiMesh configuration, unsure if this is contributing.

To Reproduce Steps to reproduce the behaviour:

  1. SSH to RT-AX92U-0480 (stock after flash) 2a. Execute 'top' and press '1' to breakdown to single CPU view and compare running processes/performance with load average displayed 2b. Execute 'uptime' to view load average, or it can be viewed from the Web UI 2c. Execute 'cat /proc/loadavg' to view the load average of the system

Expected behaviour Load average should be representative of the system's load. Given the base load on an idle RT-AX92U, my expectations would be an idle load average of <0.5.

Screenshots N/A console posted above

pixdrift commented 9 months ago

Similar issues noted in this kernel issue, but versions aren't quite the same (unless problematic patch was backported?). https://github.com/raspberrypi/linux/issues/2881

gnuton commented 9 months ago

"On Unix-like systems, including Linux, the system load is a measurement of the computational work the system is performing. This measurement is displayed as a number. A completely idle computer has a load average of 0. Each running process either using or waiting for CPU resources adds 1 to the load average. So, if your system has a load of 5, five processes are either using or waiting for the CPU."

from https://www.howtogeek.com/194642/understanding-the-load-average-on-linux-and-other-unix-like-systems/

I would not be worried about this, this is the way the linux kernel collects statistics. There may be a better way to calculate them, but as far as I know how to interpret it, I think it's not an issue.

pixdrift commented 9 months ago

Thanks @gnuton.

I understand what system load is, and also what the numbers are meant to represent. The reason for this ticket is that this build is not displaying accurate information as described in the article you linked.

A system load of ~1.9 on an idle system with 2 cores is incorrect. From the ps output there don't appear to be any processes in 'R' (runnable) or 'D' (uninterruptible sleep) states to contribute to the load.

I accept that this is unlikely an issue specific to this fork. but may possibly be a kernel driver/configuration issue.

Happy to close.

gnuton commented 9 months ago

Reopened and ready to get a patch for this. For me this an aesthetic issue and doesn't make any harm to the system performance. For this reason this gets less priority from my side. If you can find the patch of this we can apply to the kernel so that is calculated in a more correct way.