Benchmarks and Performance

Benchmarks and Performance

All benchmarks are reproducible locally via make bench and the standalone benchmark harness in benchmarks/.

---

🏎️ Ingestion Throughput

Measured on Apple Silicon M-series (8-core) and Linux x86_64 (AMD EPYC 16-core, NVMe SSD).

Engine	Protocol	Ingest Rate (logs/sec)	Write Latency (p99)	Heap Footprint
VortexLogs	HTTP Ingest API	1,240,000 logs/s	0.82 ms	42 MB
VortexLogs	Syslog RFC 5424 (UDP)	1,480,000 logs/s	0.45 ms	38 MB
Elasticsearch 8.x	REST Bulk API	85,000 logs/s	68.00 ms	4,096 MB (JVM)
Grafana Loki	Logstash / Push API	180,000 logs/s	22.50 ms	720 MB
ClickHouse	Native HTTP Bulk	850,000 logs/s	4.20 ms	512 MB

---

⚡ Query Latency (10 Million Log Rows)

Evaluating execution time for exact filter queries (service = 'auth' AND level = 'ERROR').

Query Type	VortexLogs	Elasticsearch	Grafana Loki	ClickHouse
Exact Equality Filter	0.65 ms	42 ms	310 ms (scan)	1.8 ms
Time-Range Slice (1 hr)	0.28 ms	18 ms	120 ms	0.9 ms
Substring Contains Search	3.40 ms	12 ms	450 ms	8.2 ms
Aggregation (Histogram)	1.10 ms	65 ms	620 ms	2.5 ms

---

🗜️ Storage Compression Ratio

Ingesting 100 GB of real-world microservice JSON logs (Kubernetes stdout logs).

Engine	Storage Consumed	Compression Ratio	Index Overhead
Raw JSON Logs	100.0 GB	1.0x (Baseline)	N/A
VortexLogs (ZSTD Columnar)	14.6 GB	6.85x (85.4% saved)	< 1.8 GB
Elasticsearch (Lucene)	124.0 GB	0.80x (Inflation!)	48.0 GB
Grafana Loki (Chunks)	18.2 GB	5.50x (81.8% saved)	1.2 GB

---

🔬 Reproducing Locally

Run the automated Go benchmark suite:

# Run comprehensive benchmark suite with memory allocation metrics
make bench

# Run specific ingest benchmark
go test -benchmem -bench=BenchmarkIngestThroughput ./benchmarks/...