Aidbox Scaling Strategy: Benchmarking Vertical vs. Horizontal Scaling
Health Samurai Lab: Aidbox Horizontal & Vertical Scaling
Summary
This article reports on performance benchmarks for the FHIR server Aidbox, comparing two scaling methods: concentrating CPU cores in a single instance (vertical) versus distributing them across multiple instances (horizontal). The findings suggest that while vertical scaling offers superior raw throughput (RPS), a 2x4 configuration provides the best balance of high availability and efficiency.
Details
Aidbox utilizes PostgreSQL's LISTEN/NOTIFY mechanism for inter-instance synchronization, eliminating the need for external caching services like Redis. The benchmark tested various configurations using 8 CPU cores, ranging from a single instance (1×8) to eight instances (8×1). Results showed that vertical scaling (1×8) achieved the highest Requests Per Second (RPS), demonstrating predictable and efficient growth. However, horizontal scaling efficiency degrades as more instances are used, particularly concerning memory overhead. The analysis recommends the 2x4 configuration: it maintains high availability with minimal performance loss (only ~6% RPS drop compared to 1×8) and significantly lower memory consumption than highly fragmented setups like 8×1. This suggests that for optimal balance, avoiding over-splitting is crucial.
Original content copyright by respective publishers