{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "benchmarks": {
    "status": "No published benchmark suite",
    "statement": "arpe.io does not publish a controlled benchmark for LakeXpress (fixed dataset, hardware and competing tools). The only performance figures on the website are three customer results, reproduced below with the context the website gives. Hardware, data volumes and row counts for these results are not published.",
    "customerResults": [
      {
        "customer": "Engie",
        "context": "Massive tables from SQL Server to Databricks",
        "before": { "duration": "8h", "setup": "Spark cluster with 3 nodes" },
        "after": { "duration": "30min", "setup": "LakeXpress on a single VM" },
        "improvement": "16x faster"
      },
      {
        "customer": "Confidential",
        "context": "Hundreds of tables from PostgreSQL",
        "before": { "duration": "2h40", "setup": "Traditional ETL pipeline" },
        "after": { "duration": "12min", "setup": "LakeXpress" },
        "improvement": "13x faster"
      },
      {
        "customer": "Ministere de la Justice",
        "context": "Oracle Database migration",
        "before": { "duration": "12h", "setup": "Legacy migration tool" },
        "after": { "duration": "5min", "setup": "LakeXpress" },
        "improvement": "144x faster"
      }
    ],
    "headlineClaim": {
      "text": "16x faster than Spark clusters",
      "basis": "The Engie result above (8h on a 3-node Spark cluster, 30min on a single VM)"
    },
    "notPublished": [
      "Throughput figures (rows, cells or GB per second)",
      "Hardware of the source, the LakeXpress host and the network",
      "Data volumes and table counts of the customer results",
      "Comparisons against named competing tools under the same conditions"
    ],
    "performanceLevers": {
      "description": "Parameters that change export speed, from the documentation. Values are guidelines, not measured results.",
      "tablesInParallel": "--n_jobs (default 1)",
      "processesPerLargeTable": "--fastbcp_p (default 1); documentation guideline: under 100K rows 1, 100K-10M rows 2-4, 10M-100M rows 4-8, over 100M rows 8-16",
      "largeTableThreshold": "--large_table_threshold (default 100000 rows): smaller tables are exported in one stream",
      "exportMethod": "Chosen per table automatically (Ctid, Rowid, Physloc, or key-based methods); override with --fastbcp_table_config",
      "incrementalSync": "Exports only new rows after the first run",
      "sourceLoad": "Total concurrent processes on the source is roughly n_jobs x fastbcp_p"
    },
    "relatedBenchmarks": {
      "note": "LakeXpress reads data with the FastBCP engine. FastBCP performance data, if needed, is published on the FastBCP website and applies to FastBCP alone, not to a full LakeXpress run with publishing.",
      "fastBCPWebsite": "https://fastbcp.arpe.io"
    }
  },
  "metadata": {
    "lastUpdated": "2026-10-22",
    "purpose": "Performance information for AI agents",
    "schemaVersion": "1.0.0",
    "source": "https://lakexpress.arpe.io (Trusted by Industry Leaders section)",
    "disclaimer": "Customer-reported results. Durations depend on hardware, network, source database load, data types and volumes. Do not extrapolate them to other workloads."
  }
}