Publish API benchmark
The Publish API is the primary entry point for network participants to publish and update their catalogs across the Fabric network.
The Publish API is the primary entry point for network participants to publish and update their catalogs across the Fabric network. This report presents the measured performance characteristics under realistic catalog workloads.
At a Glance
~1,000 operations/sec — Peak throughput
86–191 ms — P90 latency, 1–4 CPU range
2.52x — throughput gain, 1 CPU → 4 CPU
Methodology
The benchmarks were conducted under controlled conditions on an isolated environment to ensure consistent, repeatable results. The workload uses a standard protocol catalog dataset representative of real-world cross-domain publishing patterns.
150,000 total publish operations were executed across three configurations, based on the following fixed parameters:
Traffic Simulation
Parameter | Value |
|---|---|
Concurrent Users | 50 virtual users |
Total Requests | 50,000 payload submissions |
Payload Anatomy
Parameter | Value |
|---|---|
Payload Size | ~52 KB (average) |
Catalog Density | 2 Catalogs per request |
Resources per Catalog | 50 Resources & 10 Offers |
Total Resources per Request | 100 |
Results
Three service configurations were tested, each with increasing compute (CPU) allocation. Memory was held constant at 1 GB to isolate the effect of compute scaling.
Reading the latency columns. P90 = 90% of requests completed within that time. P95 and P99 capture the 95th and 99th percentiles — the experience of even the slowest requests.
Throughput Overview
Configuration | CPU | Throughput | P90 | P95 | P99 |
|---|---|---|---|---|---|
Standard | 1 | 383 req/s | 191 ms | 204 ms | 285 ms |
Enhanced | 2 | 605 req/s | 146 ms | 171 ms | 236 ms |
High Performance | 4 | 963 req/s | 86 ms | 106 ms | 150 ms |
Comparing the Standard (1 CPU) and High Performance (4 CPU) configurations:
Percentile | Standard | High Performance | Improvement |
|---|---|---|---|
P90 | 191 ms | 86 ms | 55% faster |
P95 | 204 ms | 106 ms | 48% faster |
P99 | 285 ms | 150 ms | 47% faster |
Latency improved consistently across all percentiles — the gains apply to every request, not just the fastest ones.
From 1 CPU to 4 CPU, throughput increased 2.52x with each doubling of CPU delivering a consistent ~1.6x gain.
Step | Gain | Efficiency |
|---|---|---|
1 CPU → 2 CPU | 1.58x | 79% |
2 CPU → 4 CPU | 1.59x | 80% |
1 CPU → 4 CPU | 2.52x | 63% |
Key Takeaways
Highly Scalable: The service scales predictably with compute resources, achieving near-linear throughput gains with ~80% efficiency per CPU doubling.
Low Latency: Even at the 99th percentile, the High Performance configuration handles complex catalog payloads in 150 ms.
Memory Efficient: All configurations ran with just 1 GB memory — even at 963 req/s, memory was not a limiting factor.
High Reliability: Zero errors across 150,000 operations demonstrates robust stability under sustained heavy load.
Benchmarking data collected on the Fabric Catalog Benchmarking environment using standard catalog payloads representative of production traffic patterns.