Linearizability
The correctness condition for concurrent objects, and how it differs from serializability (Conflict serializability and two-phase locking) and sequential...
The correctness condition for concurrent objects, and how it differs from serializability (Conflict serializability and two-phase locking) and sequential...
Adding a node should move 1/n of the keys, not all of them. One idea, and virtual nodes to fix its variance (Balls into bins and the power of two choices).
A program can be coordination-free iff it's monotone. The positive result hiding behind all the impossibility ones.
What it actually says, why it's routinely misquoted, and why latency (the "ELC") is the part that governs real designs.
How real systems dodge FLP impossibility: safety always, liveness only when the network cooperates.
One inequality generating the entire read/write consistency tradeoff space.
You need 3f+1 nodes to tolerate f liars, and that bound is tight.
No deterministic consensus in an asynchronous system with one crash fault. The rigorous theorem that CAP only gestures at.
Detecting concurrency exactly, at the cost of O(n) metadata. The tradeoff every eventually-consistent store makes.
There is no global "now"; causality is the only ordering you get. The foundational conceptual shift of the whole part.
Random placement gives max load ~log n / log log n; picking the emptier of two random bins collapses it to log log n.
Agreement over a lossy channel is impossible. Why exactly-once delivery is a lie and idempotency is the real answer.
Amdahl's rebuttal, and then the law that adds a negative term for coordination: past a point, more nodes make it slower.