Vector Clock — Junior¶

At junior level, focus on this question:

Why can't you reliably determine which of two events happened first by comparing timestamps from two different machines?

Clock skew: every machine's clock is slightly wrong¶

flowchart LR NodeA["Node A's clock:\nslightly FAST\n(+50ms drift)"] NodeB["Node B's clock:\nslightly SLOW\n(-30ms drift)"] EventA["Event on A:\ntimestamp = 10:00:00.100"] EventB["Event on B:\ntimestamp = 10:00:00.080\n(appears EARLIER, but\nmight have ACTUALLY\nhappened after A's event)"]

Every machine's clock drifts from "true" time by some small, variable amount — even with NTP (Network Time Protocol) synchronization, clocks across different machines are never perfectly aligned, and drift can be tens of milliseconds or more under normal conditions, worse during network issues. If Event A's real-world happening was caused by (came after) Event B, but A's clock happens to be running fast and B's happens to be running slow, comparing their timestamps can give you the wrong answer about which actually happened first.

Why this matters for causality, not just ordering¶

flowchart LR Write1["Write X=5 on Node A"] --> Message["A tells Node B\nabout this write"] Message --> Write2["Node B writes X=10,\nknowing about A's write\n(B's write CAUSALLY\ndepends on A's)"] Write2 --> WrongOrder["If wall-clock comparison\nsays B's timestamp is\nEARLIER than A's, a naive\nsystem might think A's\nwrite came AFTER and\nkeep the WRONG value"]

The stakes aren't just "which happened first" in an abstract sense — it's determining causal dependency: did one write happen because of information from another, such that getting the order wrong means discarding the causally-later, more-informed write in favor of an earlier one. A vector clock solves this without needing synchronized wall clocks at all — it tracks causality using nothing but per-node counters exchanged during communication.

🎓 Takeaway: wall-clock timestamps compare when something happened according to potentially-inaccurate local clocks. A vector clock instead tracks what each node knew about, and when it learned it — a fundamentally different, clock-independent basis for determining causal order.

Test yourself¶

Why doesn't NTP synchronization fully eliminate the clock-skew problem for this use case?
Walk through the "wrong order" scenario above — why does relying on wall-clock comparison risk keeping the wrong write?
What does "causal dependency" mean in this context, and why is it a different (and more useful) question than "which happened at an earlier wall-clock time"?

Continue to middle.md.