So in terms of latency/delay
- Bionic->Focal improved here
- Focal->Jammy degraded as much as we gained before
In terms of cpu cycles spent things are pretty similar per time, but that means it increased linearly with the delay which means we are slower AND consume more cpu cycles.
The programs listed with the highest consumption we see match the assumption of landscape-sysinfo and the update-motd-hwe-eol (which does the apt-config call) are the top scorer. This is based on the perf data across the full runtime which is usually pretty reliable.
Time and CPU consumption (results are rather consistent BTW):
Bionic
real 1m11.714s
us sy id wa st UTC
68 19 13 0 0 2022-03-29 13:13:25
67 19 14 0 0 2022-03-29 13:13:30
68 17 14 0 0 2022-03-29 13:13:35
59.56% landscape-sysin
16.38% swapper
7.35% lsb_release
3.77% sshd
3.22% apt-config
1.20% dpkg
0.56% dbus-daemon
0.52% systemd
Focal
real 0m44.742s
us sy id wa st UTC
67 24 8 0 0 2022-03-29 13:14:13
65 25 10 0 0 2022-03-29 13:14:18
67 24 9 0 0 2022-03-29 13:14:23
23.12% landscape-sysin
17.36% swapper
13.38% systemd-udevd
10.91% lsb_release
7.64% sshd
5.37% apt-config
1.97% dpkg
1.73% snapd
1.54% dbus-daemon
1.23% systemd
0.95% systemd-detect-
0.93% systemd-logind
0.92% gdbus
0.73% update-motd-hwe
0.69% run-parts
0.65% grep
0.61% systemd-journal
0.58% bash
0.54% find
0.52% id
Jammy
real 1m8.010s
us sy id wa st UTC
68 20 11 0 0 2022-03-29 13:15:20
68 20 11 0 0 2022-03-29 13:15:25
67 22 12 0 0 2022-03-29 13:15:30
58.52% landscape-sysin
14.82% swapper
6.28% sshd
5.72% lsb_release
3.32% apt-config
0.91% dbus-daemon
0.81% systemd
0.62% gdbus
0.54% dpkg
So in terms of latency/delay
- Bionic->Focal improved here
- Focal->Jammy degraded as much as we gained before
In terms of cpu cycles spent things are pretty similar per time, but that means it increased linearly with the delay which means we are slower AND consume more cpu cycles.
The programs listed with the highest consumption we see match the assumption of landscape-sysinfo and the update-motd-hwe-eol (which does the apt-config call) are the top scorer. This is based on the perf data across the full runtime which is usually pretty reliable.