Dynamo is the template for leaderless, eventually consistent stores: Cassandra, Riak and Voldemort descend from it. This model runs its design as a seeded discrete-event simulation: seven nodes with four tokens each on an MD5 ring, preference lists of distinct nodes, a coordinator that waits for W of N writes and R of N reads, per-server vector clocks, a sloppy quorum with hinted handoff, read repair, Merkle-tree anti-entropy and gossip membership. The page narrates the paper's Figure 3 with that cluster: five writes to a cart, a node down, a stand-in holding a hint, two clients in conflict, one merge. A research desk then asks how stale a read is. It reproduces the probabilistically bounded staleness model of Bailis et al. (2012) and checks it against the full cluster, relaxes its assumptions one at a time, tests its closed form for versions, and tests the paper's claim that an add to cart is never lost.
The machine
Seven nodes stand in a circle, each a card: for six carts, up to three versions held (colour says which version, so a conflict is two colours in one row) and the hinted replicas it keeps for others. Inside them, the ring of Figure 2 (seven tokens) and the ring the cluster uses (four tokens a node, coloured by owner), with the six carts hashed onto it. Messages are links between the cards, coloured by type. Up the table: Figure 3's clocks and who descends from whom, one stale read replica by replica, two Merkle trees, gossip round by round, the three partitioning strategies of §6.2, and the research results as blocks.
What is real. The run is a seeded discrete-event simulation of the protocol the paper describes, computed in the page by engine.js. test.mjs checks MD5 against node's own, Figure 3 clock by clock, consistent hashing's defining property, the theorem that R + W > N never reads stale across four latency models, convergence of every replica after random crashes and partitions once hinted handoff and anti-entropy run, negative controls that must fail, the closed forms and arithmetic of the Bailis paper (including two of its numbers that its own equation contradicts), and the model against the numbers that paper published: the probabilities right after a commit, Figure 4's, Table 4's t-visibilities and the N sweep. Every paper number is a fact with a verbatim quote.
Simulation · 3D
Architecture · Distributed systems
