Large-Scale System Design

Origin. Viktor Glushkov and the Institute of Cybernetics, Kyiv, 1962-1980s. Developed through the OGAS (All-State Automated System) project and related automated management system designs. Although OGAS was never fully implemented due to political opposition, the design principles became foundational for thinking about large-scale coordination systems.

Mechanism. Large-scale systems fail when they attempt to centralize information that cannot be centralized. The design insight is to separate information architecture from decision architecture: design data flows first, then locate decisions where the information required for them naturally aggregates. Hierarchical systems work when each level processes only the information relevant to its decisions and passes upward only aggregated summaries. Lateral information flows between coordinating units reduce the burden on hierarchy.

Procedure. Design large-scale coordination systems through information-first architecture: (1) Identify decisions at each level — what choices must be made centrally (parameters, constraints, strategic direction)? What must be made locally (operational details, real-time adjustments)? (2) Map information requirements — for each decision type, what information is required? Where does that information originate? (3) Design data collection at source — capture information where it is generated (enterprises, sensors, transactions) with automatic, standardized formats. Do not rely on manual reporting. (4) Design aggregation — what summary statistics does each higher level need? Aggregate upward only what is required; do not send raw data to the center. (5) Enable lateral flows — units that must coordinate should exchange information directly, not through hierarchy. Design the lateral channels explicitly. (6) Set parameters centrally, optimize locally — the center sets constraints, prices, targets; lower levels optimize within them using local information. (7) Design feedback — how do higher levels learn whether lower-level decisions are working? What monitoring, what reporting, what correction mechanisms?

Applies to. Enterprise architecture. Federated data systems. Supply chain coordination. Platform design. Any system requiring coordination without full centralization.

Limitations. The design assumes honest data input, but actors may have incentives to misreport. Any system relying on local reporting for coordination must address distortion incentives. Network effects may require critical mass; partial implementation has less than proportional value. Political resistance from those who benefit from information asymmetries (ministries, gatekeepers) can block technically sound designs. The method designs information systems but does not solve the political economy of information.