The lab starts a service that keeps one CPU busy, and adds two units that slow the boot. Reboot once after setup, so that the last boot includes them. It installs stress-ng and sysstat (with collection off), and makes the journal persistent so that task 5's evidence survives a reboot. Write answers to the files in /root/lab-answers/.

Tasks

Find from the symptoms alone which service is causing the load, and write its name to /root/lab-answers/1. Leave it running.
Make that service run at nice 15 so it no longer slows other work, in a way that still holds after the service restarts.
Find which unit slowed the last boot most, and which one was on the critical path. Write the slowest unit on line 1 of /root/lab-answers/3, and the unit on the critical path on line 2.
Turn on sysstat collection, generate load for a while, and find the busiest interval of the day (the row of sar -u with the lowest %idle). Write its time to /root/lab-answers/4.
Trigger the OOM killer with a memory hog in a throwaway service (systemd-run -p MemoryMax=100M -p MemorySwapMax=0 stress-ng --vm 1 --vm-bytes 300M --timeout 60) and find the evidence. Write the PID of the process it killed to /root/lab-answers/5.

Hints

uptime, nproc, top, systemd-cgtop, ps -o unit= -p PID.
Nice= in a drop-in (systemctl edit); ps -o ni=,comm= -C stress-ng.
systemd-analyze blame, systemd-analyze critical-chain.
systemctl enable --now sysstat (on Debian and Ubuntu, also ENABLED="true" in /etc/default/sysstat), sar -u. Collection runs every 10 minutes; running sa1 1 1 (in /usr/lib64/sa/ or /usr/lib/sysstat/) adds a row now.
journalctl -k | grep -i 'killed process', journalctl -u UNIT.