Modern Computer: Full Stack

ECE 440 · Week 5 · Lecture

One change, everything else held still

• Measure the pair, change one thing, measure the same pair the same way.
• Same supply, same cable, same card, same end event, same room: only the change differs.
• Cold start means power fully removed first; a warm reboot keeps the rails up and is a different start.
• Write the change down before you measure, so the result cannot choose the question.

ECE 440 · Week 5 · Lecture

Timing boot from the lab PC

# every console line gets a time stamp from the PC
picocom -b 115200 /dev/ttyUSB0 | ts '%.s'

• ts (from moreutils) stamps each line as it arrives.
• Boot time is the end event's stamp minus the first line's stamp, or minus the moment you apply power.
• Name which start you used; the two differ by the boot ROM and first stage.

ECE 440 · Week 5 · Lecture

What systemd-analyze does and does not see

systemd-analyze
systemd-analyze critical-chain
systemd-analyze blame

• On a board without firmware timing data it counts from the kernel's start, not from power-on.
• critical-chain shows what the end target waited for; blame sorts units by their own time.
• Use it to choose a change, not as the boot-time figure.

ECE 440 · Week 5 · Lecture

Input power, read honestly

• Put the meter between the supply and the board, so it reads everything the board draws.
• Power moves second to second: record over a window and say average or peak.
• Read in the same state each time, for example idle at the prompt for a fixed wait.
• Write the unit the meter shows next to every number you copy.

ECE 440 · Week 5 · Lecture

Each chart starts at zero

import matplotlib.pyplot as plt

fig, ax = plt.subplots()
ax.bar(["before", "after"], [before, after])
ax.set_ylim(bottom=0)
ax.set_ylabel("boot time (s)")
fig.savefig("boot.png")

A cut baseline makes a small change look large. One quantity per chart, unit on the axis title.