Where the Milliseconds Actually Go in a Round Trip

[Shadow Trader]

Round-trip latency is not a single number. A market order passes through several stages, and each one adds delay independently: the trader's local machine, the network path to the broker, the broker's trading server, any risk or dealing logic, the link to a liquidity provider, and the confirmation traveling back.

Ping tests only measure the network stage, and often only to a gateway rather than the actual trade server. A fast ping can still coexist with slow fills if server-side processing is queued or if checks at the liquidity provider add a hold time before acceptance.

Separating the stages is the practical way to reason about it. Timestamps in the platform journal, compared against the time reported on the fill, can reveal how much of the delay occurred before the order reached the server versus after. Repeated samples matter more than single readings, since latency tends to be spiky around news and rollover.

For targets measured in a few pips, a consistent 80 milliseconds can be easier to plan around than a low average with frequent large outliers. Variance, not just the mean, deserves tracking.

[FXS HongKong Scalper]

Good breakdown. The point about ping only measuring the network leg is what most people miss. We once had a VPS with 2 ms ping to the broker and fills that took 150 ms on average. The delay was on the broker side, probably a risk check for orders above a certain size.

To separate the stages, we log three times in the EA: the moment OrderSend is called, the moment the result returns, and the server time on the deal. The difference between the first two is round-trip, and comparing the deal's server time with the request shows how much of that delay is on the server side, as long as the clocks are synced.