Ping to the broker is not latency to the market
Posted: Fri Oct 30, 2026 12:21 pm
Ping to the broker is not latency to the market
A VPS provider advertises 2 milliseconds to the broker's data centre. Impressive. The question is what happens after the broker's gateway.
An order passes through the broker's trading server, then a bridge, then one or several liquidity providers, and the confirmation returns the same way. Each step adds delay. The ping measures only the first.
Some brokers run their servers in the same building as their LPs. Others route through distant cities. The number that matters is the round trip from click to fill confirmation, and it can be ten or twenty times the ping.
How to measure it: place small test orders, record the time the order was sent and the time the fill confirmation arrived, and repeat a hundred times. Then compute the median and the 95th percentile.
The median shows typical speed. The percentile shows the tail, where scalping losses actually accumulate. A broker with a lower median but a fat tail can be worse than one with a higher, steadier median.
One more consideration is that execution speed often varies by time of day. A test at 03:00 says little about 13:30.
Another factor is the type of order. Market orders, pending orders, and modifications can each take different paths and different times.
Having this data lets a scalper choose between brokers on evidence, and decide whether the strategy's time horizon is even compatible with the delays.
A simple extra measurement is the gap between the order's server time and the fill time shown in the deal history, which isolates the broker-side part of the delay. Subtracting that from the total round trip leaves the network portion. If the network part is small and the broker part is large, a faster VPS will not help, and the only remedy is a different broker.
How different was the measured round trip from the advertised ping, and how fat was the tail at the busiest hours?
A VPS provider advertises 2 milliseconds to the broker's data centre. Impressive. The question is what happens after the broker's gateway.
An order passes through the broker's trading server, then a bridge, then one or several liquidity providers, and the confirmation returns the same way. Each step adds delay. The ping measures only the first.
Some brokers run their servers in the same building as their LPs. Others route through distant cities. The number that matters is the round trip from click to fill confirmation, and it can be ten or twenty times the ping.
How to measure it: place small test orders, record the time the order was sent and the time the fill confirmation arrived, and repeat a hundred times. Then compute the median and the 95th percentile.
The median shows typical speed. The percentile shows the tail, where scalping losses actually accumulate. A broker with a lower median but a fat tail can be worse than one with a higher, steadier median.
One more consideration is that execution speed often varies by time of day. A test at 03:00 says little about 13:30.
Another factor is the type of order. Market orders, pending orders, and modifications can each take different paths and different times.
Having this data lets a scalper choose between brokers on evidence, and decide whether the strategy's time horizon is even compatible with the delays.
A simple extra measurement is the gap between the order's server time and the fill time shown in the deal history, which isolates the broker-side part of the delay. Subtracting that from the total round trip leaves the network portion. If the network part is small and the broker part is large, a faster VPS will not help, and the only remedy is a different broker.
How different was the measured round trip from the advertised ping, and how fat was the tail at the busiest hours?