austere turtle
.....Is my debian/KDE desktop supposed to have 2k processes, handling 20k-100k context switches per second?
I'm trying to track down some input lag that's happening everywhere, and so far, the most likely candidate is how some processes are regularly getting starved of CPU time, despite the fact that my CPU has virtually no utilization.
Question: Is 100k - 200k us of delta between requested sleep time and actual sleep time normal?
I wish I could share these netdata graphs I'm looking at as a link to a webpage, but there's not an obvious way to do that, so maybe I'll look that up later
System Active Processes
Context Switches:
"Idle Jitter" or the difference between requested sleep time and actual sleep time.
Interrupts:
"System Interrupts":
Soft IRQ latency seems comparable to hardware IRQ frequency (5-40ms), so maybe the volume of interrupts isn't that bad.....?
Looks like disk writes were elevated:
.....removed VFS Files....?
...Under networking, "ECN"?
Bandwidth and packet monitoring:
(@cerulean merlin, you might appreciate seeing what I'm doing with the debian rig here...)
austere turtle
I think those graphs are all the insight I have for today. I don't have any leads off the top of my head, but I do want to figure out what all these thousands of processes are...
Only 450-odd processes according to ps. Still a lot, but it's mostly stuff I recognize.
molten sluice
VM or bare metal? is this after you were going to compile a new kernel?
I'm assuming that it's the same machine that was failing Fedora's install
austere turtle
Yeah, this is my desktop, running bare-metal. I removed the bad stick of RAM, so my desktop should be less cursed now.
The new kernel compile only added a driver for the Corsair Commander Pro. If it's something as simple as switching schedulers, then that's a thing I can do. I just don't know that is the solution yet. I just don't know what these graphs actually mean yet.
austere turtle
I gave up and decided to try removing the xorg wacom input driver, as suggested in https://forums.debian.net/viewtopic.php?t=65448 . After rebooting, first signs show promise, but idk why it'd help....
Maybe the reboot was part of the fix? System active processes is down by 400-500 for now...
Context switches does not appear to be currently impacted
idle jitter is massively improved, currently around 100-200us (vs being off by tens of thousands)
Can't comment about system interrupts
Disk activity appears to have been some unrelated load, it calmed down on its own, and wasn't affected by the changes I made.
ECN appears unaffected by changes.
Network appears to have calmed down on its own, outside of my changes.
So, if the fix sticks, wacom xorg driver is bad, and causes idle jitter measurable in ms (possibly only if there's no wacom device installed).
molten sluice
Only 450-odd processes according to `ps`. Still a lot, but it's mostly stuff I r...
I'm 557 on a fairly utilized Ubuntu desktop
ps -ef | grep "" -c
austere turtle
Does that grep call count matching lines? I guess that's easier than remembering wc -l ....
molten sluice
Does that `grep` call count matching lines? I guess that's easier than rememberi...
yes
I was trying to remember, thought it was `cat' something, grep is a close second
austere turtle
Didn't even think about using grep that way, but it's a cool trick. Yeah, cat is for combining a bunch of files together, grep for finding stuff, and wc for counting words/lines/etc.
But, if the count is supposed to be < 1000, that's cool. But then why is netdata reporting such a massively larger number than ps?
molten sluice
I use grep that way to help figure out how many instances of an error in a log or something