Bagaskoro Saputro
We present an empirical performance evaluation of SILM, a national-scale music-royalty administration platform prototyped for LMKN, Indonesia's collective rights management agency, implemented as a chain of ten event-driven Go microservices connected through an in-memory publish/subscribe bus (Apache Kafka in the production blueprint). The study contributes a Dapper-style trace-per-event instrumentation yielding per-stage latency distributions, a five-point throughput sweep from 100 to 10,000 play events used to locate the operating point and the degradation knee, and a money-conservation and correctness suite. At every tested burst scale up to 10,000 events the pipeline delivers 100\% event delivery and exact money conservation, while median end-to-end (E2E) latency grows approximately 17-fold (0.81 s at 100 events to 13.7 s at 10,000 events); per-stage spans attribute 85.0\% of E2E average latency at the largest scale to a single cross-service queueing stage. A 30-second sustained-load soak at approximately 1,042 events/s exposes the single-consumer ceiling: the bounded subscriber queue overflows in its tail, dropping 7,097 of 31,255 submitted plays (22.7\%), the first measured reliability failure of the platform. All findings are compared against recent published results on tail latency, bottleneck attribution, and channel sizing in event-driven architectures.