Broker VPS Directory / Latency dataset
Real TCP round-trip times from our VPS regions to 4,092 live MT4/MT5 broker servers across 2,207 brokers, measured rather than estimated, every row dated. Free to download and reuse with attribution.
Download
broker-latency.csv 4,092 rows · CC BY 4.0
Columns: broker_server, broker_company, platform, vps_city, vps_country, measured_tcp_ms, measured_date.
| Broker servers measured | 4,092 |
| Brokers covered | 2,207 |
| Region/server pairings measured in total | 16,617 |
| Median best-region round-trip | 2.0 ms |
| Measurement window | 2026-08-09 → 2026-09-01 |
| Licence | CC BY 4.0 |
Every location we can actually sell, with what we measured from it. Sorted by how many broker servers it is the closest match for.
| City | Country | Servers | Fastest | Median | Slowest |
|---|---|---|---|---|---|
| London | United Kingdom | 1,277 | 0.3 ms | 1.8 ms | 56.6 ms |
| Amsterdam | Netherlands | 571 | 0.3 ms | 3.0 ms | 1029.0 ms |
| Hong Kong | Hong Kong | 499 | 0.3 ms | 2.4 ms | 1016.0 ms |
| Frankfurt | Germany | 428 | 0.5 ms | 1.4 ms | 27.9 ms |
| Singapore | Singapore | 403 | 0.3 ms | 1.6 ms | 189.7 ms |
| Toronto | Canada | 161 | 0.9 ms | 1.5 ms | 108.8 ms |
| New Jersey | United States | 112 | 0.8 ms | 3.2 ms | 82.9 ms |
| Montreal | Canada | 104 | 0.6 ms | 1.4 ms | 11.5 ms |
| Dublin | Ireland | 100 | 2.4 ms | 20.4 ms | 122.6 ms |
| Seoul | South Korea | 100 | 0.3 ms | 1.1 ms | 109.3 ms |
| Ashburn | United States | 74 | 1.1 ms | 1.8 ms | 86.6 ms |
| Delhi NCR | India | 61 | 17.5 ms | 19.3 ms | 1024.0 ms |
| Tel Aviv | Israel | 43 | 1.1 ms | 100.8 ms | 163.1 ms |
| Paris | France | 42 | 0.3 ms | 2.3 ms | 12.8 ms |
| Johannesburg | South Africa | 37 | 0.5 ms | 31.5 ms | 175.2 ms |
| Warsaw | Poland | 28 | 0.3 ms | 3.3 ms | 67.2 ms |
| Dallas | United States | 14 | 0.5 ms | 0.6 ms | 4.9 ms |
| Chicago | United States | 8 | 1.2 ms | 6.4 ms | 22.2 ms |
| Los Angeles | United States | 7 | 0.5 ms | 1.2 ms | 33.5 ms |
| Silicon Valley | United States | 7 | 2.5 ms | 12.8 ms | 101.4 ms |
| Seattle | United States | 7 | 1.3 ms | 2.0 ms | 205.4 ms |
| Portland | United States | 6 | 3.9 ms | 9.2 ms | 11.3 ms |
| Miami | United States | 2 | 1.8 ms | 36.7 ms | 36.7 ms |
| Cape Town | South Africa | 1 | 53.0 ms | 53.0 ms | 53.0 ms |
Each figure is a TCP handshake from a probe in the region we would sell, to the broker server's own published host and port — not an ICMP ping, and not a ping to our own network.
Validation. We compared this method against MetaTrader's own reported round-trip on 10 live MT5 terminals running on the same machine as the probe: 8 of 10 agreed within 10 ms, median difference +1.3 ms.
Known failure mode, stated plainly. MT4 and MT5 frequently connect to an access server which then redirects the terminal to a different trade server. Where a broker's access point is far from its matching engine, our figure reads worse than the connection a trader actually gets — we observed one access point measuring 162 ms whose real terminal ping was 8 ms. We do not correct for this, because only a live terminal on that account can reveal it.
Error direction is consistent. The TCP figure reads higher than real terminal ping, effectively never meaningfully lower. Every number here is therefore a conservative upper bound — it can undersell our own product but cannot overpromise.
What is excluded. Servers whose published address is a private or reserved IP (no real endpoint can exist there), and servers that stopped answering across a three-pass sweep.
Licensed CC BY 4.0. Attribution required — suggested form:
metatraderVPS (2026). Measured MetaTrader Broker Latency Dataset. Retrieved from https://metatrader-vps.com/vps-hosting/latency-dataset/
Yes. It is published under CC BY 4.0: use it commercially, modify it, redistribute it, provided you credit metatrader-vps.com and link back to this page. If you need a different arrangement, ask.
Those figures are almost always a ping inside the provider's own network, which says nothing about your broker. Every row here is a TCP round-trip from the region we would actually sell you to the broker's own published host and port, with the date it was taken.
The download gives the best-region figure per broker server -- the same number already shown publicly on that server's own page -- plus aggregates. The complete 16,617-pairing matrix took a long unattended sweep to build and is not published. The aggregates below are computed from all of it.
Periodically, because network routes drift and brokers relocate servers. Every row carries its own measurement date rather than a single dataset-wide timestamp, so you can judge staleness per broker instead of trusting one number. Current rows span 2026-08-09 to 2026-09-01.
Broker server locations are gathered via public network lookups and provided for informational purposes only. Latency figures are measured TCP round-trips to each broker's own published address on the date shown, not guarantees of future performance. Not affiliated with any broker listed.
