Performance benchmarks at booking scale

Benchmark at a Glance

Dataset: 60M records (100k properties × 10 room types × 60 days)
Data generation: 4 segments × 25k properties × 10 room types × 60 days
Hardware: 4-core i7 laptop, 16 GB RAM
Memory usage: ~4.1 GB for 60M records
Load time: 3 seconds from compressed snapshot

Roomzin Read Performance

Regular search · 1000 connections · 20k requests · 1s duration

Metric	Value
Min	0.001 ms
P50	0.490 ms
P95	1.472 ms
P99	3.313 ms
P99.9	3.842 ms
Max	4.557 ms
Requests per second	20,000

* Benchmark run with the official roomzin-bench tool

Roomzin Spike Performance

Simultaneous burst · 100 connections · 2,000 requests · 0s ramp-up

Spike: All 2,000 requests are sent at exactly the same time with no delay between them. This simulates a sudden traffic surge where all requests arrive simultaneously.

Metric	Value
Min	1.692 ms
P50	23.970 ms
P95	46.021 ms
P99	47.874 ms
P99.9	48.077 ms
Max	48.087 ms
Requests	2,000
Concurrent connections	100

RzProxy Read Performance

HTTP proxy · 200 connections · 2s duration

Metric	Value
Min	0.20 ms
P50	6.23 ms
P95	11.53 ms
P99	13.81 ms
Mean	6.29 ms
Total requests	43,692
Achieved RPS	21,846

* Benchmark run with the official roomzin-bench tool

Snapshot Performance

Compact snapshots enable fast saves and reloads. Compressed snapshots are ~92 MB for 60M records.

Records	Load Time	Raw CSV Size	Snapshot (compressed)	Memory Usage
60M records	3,041 ms	5.3 GB	92 MB	~4.1 GB

The in-memory footprint is 4.1 GB, which is ~77% of the raw CSV data size (5.3 GB)

In Production

Multiple read replicas (follower nodes) provide better results
RzProxy independent scaling provides higher TPS for HTTP clients
More cores provide better parallelism. Segments are processed in parallel.

Benchmark tool: github.com/m-javani/roomzin-bench — reproduce these results or test your own setup.