How to measure real execution latency instead of trusting the ping number
A ping of 8 ms to the broker server feels reassuring, but ping measures the network leg only. Execution latency includes the platform, the broker's bridge, risk checks, the liquidity provider response, and the trip back. It can be fifty times larger than the ping.
A clean way to measure it is simple. Log the timestamp when the order is sent from code or the click, then log the timestamp of the fill confirmation. Subtract. Do it for hundreds of orders across different hours.
The distribution matters more than the average. A broker with a 60 ms median and a handful of 900 ms outliers can hurt a scalper more than one with a steady 150 ms. The outliers tend to cluster around news and session opens, exactly when precision counts.
Also compare latency by order type. Market orders, pending order triggers, and modifications often travel through different paths.
Keep the log. It becomes a powerful argument in any support ticket.
What median and worst-case latency has your logging turned up, and at what times of day?
How to measure real execution latency instead of trusting the ping number
-
Amanhasnoname
- Posts: 2060
- Joined: Fri Oct 30, 2026 9:45 am
Re: How to measure real execution latency instead of trusting the ping number
I used to quote my VPS ping of eleven milliseconds as if it described execution. Once I wrote the send time from code and the deal time from the trade transaction into a file, the picture changed. Over six hundred and eighteen market orders the median round trip was one hundred and forty-two milliseconds. The average sat near two hundred and seven because of a fat tail: twenty-three fills took longer than eight hundred milliseconds, and fourteen of those landed in the first twenty minutes of London or around a red-folder print.
Pending stops were slower. Their median from trigger to fill was two hundred and ninety milliseconds, and some stop modifications sat for half a second before the server confirmed. Market orders in the quiet Asian window were the steadiest, usually between one hundred and one hundred and sixty milliseconds.
I keep that log and send a slice of it whenever support tells me the connection looks fine. The ping number has barely moved. The outliers have, and those are what I actually care about.
Pending stops were slower. Their median from trigger to fill was two hundred and ninety milliseconds, and some stop modifications sat for half a second before the server confirmed. Market orders in the quiet Asian window were the steadiest, usually between one hundred and one hundred and sixty milliseconds.
I keep that log and send a slice of it whenever support tells me the connection looks fine. The ping number has barely moved. The outliers have, and those are what I actually care about.