Why FIFO Isn't Enough

In a group chat, Bob asks "what time?" and Alice replies "noon". Charlie joins the conversation. If Charlie's network delivers Alice's "noon" before Bob's question, Charlie sees a non-sequitur — an answer to a question he hasn't seen yet.

FIFO doesn't help here because Bob and Alice are different senders. FIFO orders only same-sender messages.

What we need is causal delivery: whenever $\text{send}(m_1)\to \text{send}(m_2)$ and both target the same destination, $m_1$ must be delivered before $m_2$. The arrow $\to$ is Lamport's happened-before — covering all cause-and-effect chains, including those going through intermediaries.

BSS — The Algorithm

Birman, Schiper, Stephenson (1991). Each process maintains a vector clock $V_i$. On send to $P_j$, increment $V_i[i]$ and piggyback $V_i$ as $V_m$ on the message.

On receive of $m$ from $P_j$ at $P_i$, deliver only when BOTH:

(a) $V_m[j]=V_i[j]+1$ — this is the next expected message from $P_j$ (FIFO from $P_j$).

(b) $\forall k\ne j:V_m[k]\le V_i[k]$ — all causally preceding messages from other senders have arrived.

If either fails, buffer the message. Re-check when other messages arrive.

After delivery, $V_i[j]:=V_i[j]+1$.

Why The Two Conditions

Condition (a) is FIFO — don't skip messages from the same sender.

Condition (b) is the causal part. The sender's vector $V_m$ reflects which other senders' messages the sender knew about. If $V_m[k]>V_i[k]$, the sender saw a more recent message from $P_k$ than $P_i$ has — meaning $P_k$ sent something to $P_i$ that hasn't yet arrived. Deliver this message and you'd be delivering effect before cause.

Order Strengths In One Picture

| Property | Orders | Concurrent messages | Cost | |---|---|---|---| | FIFO | same-sender pairs | unordered | very cheap (just sequence numbers) | | Causal | causally-related pairs | unordered | moderate (vector clock per message) | | Total | ALL pairs | also ordered (consistent) | heavy (needs consensus) |

State-machine replication (Raft, Paxos) needs total order. Group chat usually wants causal. TCP gives FIFO.

What You Walk In Carrying

Definition of causal order. Why FIFO is strictly weaker. BSS two conditions and their meaning. Strength order: Total ≻ Causal ≻ FIFO. Use cases: replicated DBs, group communication, distributed debugging.