Week 10 Session: Measuring a kernel

Take your own numbers off a prepared machine, then work out what a whole operating system would cost your target

3 May 2027

Nobody is booting a kernel for their own target in an hour, so this week you measure a rig that already works and reason from it.

You are given a browser-hosted machine running a real operating system. Measure, rather than read:

• Cold start. First byte to usable prompt. Then do it again with a warm cache and note the difference.
• Size. What was downloaded, and what is resident afterwards.
• Throughput. Something CPU-bound, against the same work on your host. Get a ratio.
• Latency. Typing. It is subjective and it matters more than the ratio.
• The thing that breaks. Use it as you would a real machine for five minutes and find the first thing that is not there. Copy and paste between the page and the guest. Resize the window. Play a sound. Open a second core’s worth of work. Unplug the network. Reload the tab and see what survived.

Then the estimate. For your target: what would this approach cost in download, in start-up, in throughput — and, honestly, what would it fix? A kernel solves processes, filesystems and sockets in one move. If your target needed all three, this may be cheaper than five weeks of shims. If it needed none of them, you are paying tens of megabytes for nothing.

Bring one row for your evidence log with a real number in it, and a sentence saying whether this option is still live for you. “Eliminated, on these numbers” is a good outcome and you will be asked to defend it in week 12.