teleo-codex/domains/internet-finance/aimd-converges-to-fair-resource-allocation-without-global-coordination-through-local-congestion-signals.md

type	domain	description	confidence	source	created	secondary_domains
claim	internet-finance	AIMD algorithm achieves provably fair and stable distributed resource allocation using only local congestion feedback	proven	Corless, King, Shorten, Wirth (SIAM 2016) - AIMD Dynamics and Distributed Resource Allocation	2026-03-11	mechanisms critical-systems

AIMD converges to fair resource allocation without global coordination through local congestion signals

Additive Increase Multiplicative Decrease (AIMD) is a distributed resource allocation algorithm that provably converges to fair and stable resource sharing among competing agents without requiring centralized control or global information. The algorithm operates through two simple rules: when no congestion is detected, each agent increases its resource consumption additively (rate += α); when congestion is detected, each agent decreases multiplicatively (rate *= β, where 0 < β < 1).

The SIAM monograph by Corless et al. demonstrates that AIMD is mathematically guaranteed to converge to equal sharing of available capacity regardless of the number of agents or specific parameter values chosen. Each agent only needs to observe local congestion signals—no knowledge of other agents, total capacity, or system-wide state is required. This makes AIMD the most widely deployed distributed resource allocation mechanism, originally developed for TCP congestion control and now applicable to computing, energy grids, and other domains.

The key theoretical properties are:

• Fairness: Converges to equal per-agent allocation of available resources
• Stability: System reaches equilibrium regardless of agent count or parameters
• Locality: Decisions based only on local congestion observation, not global state
• Simplicity: Two-rule algorithm with no optimization or prediction required

These properties make AIMD particularly valuable for systems with variable load and expensive coordination costs, where reactive control based on observed system state outperforms predictive models or centralized optimization.

Evidence

The SIAM monograph provides formal proofs of convergence and stability across three application domains:

1. Internet congestion control (TCP)
2. Smart grid energy allocation
3. Distributed computing resource management

The algorithm's widespread deployment in TCP—handling the majority of internet traffic for decades—provides empirical validation at planetary scale.

Application to Pipeline Architecture

For systems with expensive evaluation compute and variable extraction load, AIMD provides a principled scaling mechanism: when the evaluation queue shrinks (no congestion), increase extraction workers by 1 per cycle; when the queue grows (congestion detected), halve extraction workers. This approach avoids the complexity of load prediction, arrival modeling, or solving optimization problems while guaranteeing convergence to sustainable throughput.

---

Relevant Notes:

• optimal governance requires mixing mechanisms because different decisions have different manipulation risk profiles
• coordination mechanisms
• designing coordination rules is categorically different from designing coordination outcomes as nine intellectual traditions independently confirm

Topics:

• domains/internet-finance/_map
• core/mechanisms/_map
• foundations/critical-systems/_map