Performance

Real workloads. Real networks. Honest baselines.

Every benchmark on this page links to its full methodology — workload, network shape, comparison baseline. We publish parity results too, because honesty is the differentiator.

2.5x
WAN throughput
On 50 Mbit / 170 ms RTT links vs a typical proxy baseline
139x
SQL Server bulk loads
Cloud-region bulk load vs native client tooling
70x
MySQL single-row
Single-row workloads vs native client tooling

Network shape methodology

Network shape is the dominant variable in distributed-database performance. Every benchmark documents the RTT, bandwidth, and loss profile it ran on. Reproducible harnesses are linked from each result.

  • RTT, bandwidth, and loss simulated with deterministic shapers
  • WAN reference shape: 50 Mbit, 170 ms RTT, 0.1% loss
  • LAN reference shape: 1 Gbit, sub-1 ms RTT, 0% loss
  • Each scenario runs against a comparison baseline — never naked numbers
  • Source code and raw output published alongside each report

Workload methodology

Application workloads behave differently. We measure three shapes that cover most production traffic — and report them separately so you can pick the one closest to yours.

  • Single-row — high-rate small queries with frequent round trips
  • Pool-borrow — workloads dominated by connection acquisition and reuse
  • Bulk — large transfers (BCP-class loads, COPY, mysqldump-style restores)
  • Each workload reports median and high-percentile throughput separately

Headline results

Every row links to a full methodology section.

WorkloadNetwork shapeResultBaseline
WAN throughput50 Mbit / 170 ms RTT~2.5xTypical proxy baseline
Pool-borrow heavyWAN~2.3xTypical proxy baseline
LAN baseline1 Gbit / sub-1 msParityTypical proxy baseline
SQL Server bulk (cloud)Cloud cross-AZUp to 139xNative client tooling
MySQL single-rowWANUp to 70xNative client tooling

Numbers reflect representative results from the harnesses linked above. Your workload, your network, your numbers — see them on a 30-minute demo.

Where DBWarp doesn't win

On low-latency LAN connections, DBWarp runs at parity — there's no penalty for keeping it in the data path, and there's no headline number to claim. We publish parity results because the alternative is dishonest. Performance gains scale with the network challenge: the further the database, the bigger the win.

Want the numbers on your workload?

A 30-minute demo with your real network shape, your real database, your real queries.