Experimental Evaluation of Cross-Domain Data Synchronization Architectures Under Consistency and Failure Constraints

Abstract

Domain-oriented architectures improve ownership boundaries in large enterprise systems, but operational workflows frequently require data owned by neighboring domains. This creates a recurring design problem in which teams must balance freshness, availability, and operational isolation when sharing data across domain boundaries. This paper empirically evaluates cross-domain synchronization behavior in a containerized retail reference system implementing direct application programming interface (API) consumption, event-driven projection, batchoriented replication, and hybrid routing. At 100 concurrent users, increasing injected Client API delay from 0 to 2000 ms increased Orders-service p95 latency from 497 to 2159 ms and reduced throughput from 345.5 to 48.5 requests/s, demonstrating the downstream cost of synchronous dependency. Across ten independent trials, an event-driven projection recovered a 10,000-event backlog in 12.13 s on average, corresponding to a mean catch-up rate of 824.7 events/s. Protocol inspection also identified two benchmark defects that invalidated the intended batch-interval and freshness service-level agreement (SLA) comparisons: the interval configuration was not applied to the batch worker, and the SLA phase reused previously synchronized state. Those outcomes are analyzed as benchmark-design findings rather than reported as comparative performance evidence. The results demonstrate that credible evaluation of cross-domain synchronization requires both architecture-level measurement and strict control over workload state, treatment configuration, and update-to-read timing.