Ada Audits the Throughput and Reliability Numbers
Ada re-derives this chapter’s own numbers step by step, at full precision
ADA · CALCULATION AUDIT
Ada Audits the Throughput and Reliability Numbers
A 50-robot order fleet clears one order every 45 s, so each robot handles 80 orders/hr and the fleet tops out near 4000 orders/hr — short of the 10,000/hr peak — while a 5 s broker crash on the transient path can drop $250,000 of orders. Persistence and manual acknowledgment each buy that safety by spending throughput or latency. This audit re-derives every ceiling and asks whether the ack-versus-persistence call is philosophy, or simply the ledger these numbers spell out.
Companion to the chapter AMQP Reliability Patterns — every number here comes from that chapter.
Every reliability knob in this chapter — persistence, manual ack, prefetch — buys safety by spending throughput or latency, and the numbers say exactly how much. Each figure below is one the chapter already stated; I only carry the division through.
1. Latency sets a hard per-connection throughput ceiling.
If a message cannot clear until its round-trip of L milliseconds completes, one connection tops out at T = 1000 / L messages per second:
That 20× gap is the cost of writing each persistent message to disk instead of holding it in memory.
2. Manual ack is cheap when processing dominates; the fleet ceiling is set by processing time.
A 45 s order plus a ~50 ms ack round-trip wastes only 0.05 / 45.05 of the time, and each robot clears 3600 / 45 orders per hour:
| Reliability question | Arithmetic shown | Audit result |
| Persistent throughput ceiling (10 ms disk) | 1000 / 10 | 100 msg/s per connection |
| Transient throughput ceiling (0.5 ms memory) | 1000 / 0.5 | 2000 msg/s per connection |
| Persistence throughput penalty | 2000 / 100 | 20× slower |
| Per-robot order rate | 3600 / 45 | 80 orders/hr |
| 50-robot fleet capacity | 50 × 80 | 4000 orders/hr (below 10,000 peak) |
| Manual-ack overhead per message | 0.05 / 45.05 | 0.11% (negligible) |
| Lost orders per 5 s crash (transient) | 50 × 5 × $1000 | $250,000 |
What this means for your design: the ack-versus-persistence decision is not philosophy, it is this ledger. Persistence cuts the per-connection ceiling twentyfold, so route only the messages that must survive — the orders worth $250,000 a crash — through the durable path and keep high-volume telemetry transient. Manual ack costs 0.11% when processing dominates, so use it freely for critical work. And note the 4000 orders/hr fleet ceiling comes from the 45 s processing time, not the broker: to reach the 10,000/hr peak you scale consumers, not reliability knobs.
Every number above is taken from the chapter’s own examples and re-derived step by step.