Interviewers rarely fail candidates for not knowing a state name. They fail answers that skip the model, describe the protocol in vocabulary only, or cannot say how the claim would be checked. This topic gives a fixed structure for any coherence answer, a two-minute drawing routine for a transaction, a four-step way to explain a race, and a rubric you can apply to your own answers. It adds no new protocol content: every worked case is a trace from Parts 2 and 4, linked rather than redrawn.
Lesson | After it you can... |
|---|---|
Structure an answer as Assumption, Mechanism, Trade-off, Evidence, and tell a weak answer from a strong one. | |
Draw lanes, a starting snapshot, numbered messages and a state/value table, and check SWMR and the data-value invariant at the end. | |
Name both transactions, the serialization point, who waits or retries, and why the newest value is never lost. | |
Score an answer from 0 to 4 and know what is missing for the next level. |
Suggested order: framework, drawing, race, rubric. Then practise on the question bank in this Part, grading each attempt with the rubric. Keep the course model and the revision sheet open: a strong answer names the model before it names a message.
Assumption. The same question has different correct traces under an ordered snooping bus, a directory with acks at the home, or a directory with acks at the requester. Without the assumptions, the interviewer cannot tell whether the mechanism is right for the system being discussed, and the answer cannot reach the top of the rubric.