Comment 107 for bug 1208993

Alright, I have run some more tests:

I copy a file in my (unencrypted, ext4, on root partition) homedir with:

dd if=testfile_8gb of=testfile_out status=progress

(I made the testfile using dd if=/dev/zero of=testfile_8gb bs=1M count=8000 status=progress)

While running this Xorg/Gnome3/Mouse/Any UI application becomes unresponsive for seconds at a time, with intervals of a few seconds where everything is responsive. The write speed of dd often starts somewhere between 200 and 100 mb/s and then dips all the way to around 35, after which the dd process completely locks up. (I think this is because the process gets stuck into some kind of iowait state, I deduced this from some quick peeks at KSysGuard, citation needed). After a few minutes the dd process suddenly finishes and the UI becomes responsive again. Copying the file through nautilus has similar behaviour.

Disabling intel_pstate via the kernel boot parameters (intel_pstate=disabled) did not have any effect (except for indeed disabling intel_pstate).

Doing some tweaks to the vm dirty bytes settings had no significant effect (making the mouselag more of a jitter, but still unusable).

Disabling the swapfile (19.10 has a swapfile instead of swap partition by default) however completely mitigated the issue!

If some people are willing to try to reproduce this we might be able to get something going:

- Open some Youtube video in Firefox (something that lets you see if the UI lags)
- next to that open two terminals, in one terminal start the dd process as described above
- The ui should be lagging now, mouse doesn't move for seconds at a time, youtube video severaly lags/hangs/jitters
- in the second unused terminal run 'sudo swapoff /swapfile'
- For me the lag immediately stops
- Killing the dd process/letting it finish and then starting a new one works perfectly, much higher writespeeds (and more stable)
- enable swag again with 'sudo swapon /swapfile'
- Lag returns completely

I am really curious if this works for others. Maybe this is an entirely different issue, for which I will file a bugreport if nobody can reproduce this.

Thanks in advance!