Benchmarks & Cost

New Smart Core: The latest engine (December 2025) now runs comfortably on 100MB RAM (0.5 vCPU).

We verified these results in a production dataset (23,000 Records / 8 Concurrent Pipelines) on a constrained $5.00/mo VPS.

1. Throughput Capacity

Per Minute

42k

Rows

Per Hour

2.5M

Rows

Per Day

60M

Capacity

Processing Time (Lower is Better)

Task: Sync 14,000 complex JSON objects to Postgres.

ApiTap (Smart Core)20s

✓ Success

❌Airbyte (Java) - Failed (OOM)∞

✗ Out of Memory

2. Cost Efficiency Analysis

Comparing ApiTap (Self-Hosted on $5/mo VPS) vs Airbyte Cloud (API Pricing).

Metric	ApiTap (Self-Hosted)	Airbyte Cloud (API)	Conclusion
Cost Per Million Rows	$0.003	$15.00	~5,000x Cheaper 🎉
Cost for 1 Billion Rows	$2.70	$15,000.00	Save $14,997
Monthly Minimum	$5.00 (VPS Capacity)	$10.00 (Credits)	-
Infrastructure	0.5 vCPU / 100MB RAM	Managed Cloud	-

✅Production Stability: 18 Hours Continuous Running

Real Production Test: Running 5 pipelines every minute for 18+ hours on minimal resources.

CPU Usage:~5%

Memory Usage:30-70 MiB

Resource Limit:0.5 vCPU

Pipeline Performance (per execution):

✓github_apache:18.1s(2839 records)

✓github_koteka:17.4s(2839 records)

✓github_supabase:21.1s(2839 records)

✓github_risingwave:21.0s(2839 records)

✓github_aduhai:19.1s(2839 records)

Result: Memory stable at ~100MB with zero restarts required. Running 8 parallel pipelines effortlessly on 0.5 vCPU!

3. Why is it so efficient?

Zero-Buffered Streaming

Unlike traditional ETL tools that load thousands of records into memory arrays (causing standard runtimes like Node or Java to bloat), ApiTap creates a "stream pipe":

1
Fetch: Pulls 1 page (50 items).
2
Write: Instantly sends to Postgres using Arrow vectors.
3
Drop: Memory is cleared immediately (< 1ms).

Smart Core Tuning

When running on low-memory servers (detected via Cgroups):

1
Jemalloc (narenas:1): Forces aggressive memory cleanup to prevent fragmentation.
2
Tokio Worker Control: Limits thread overhead to fit within 256MB constraints.