Windows timer resolution and the 15 millisecond granularity
Posted: Sat Oct 31, 2026 2:24 pm
Windows schedules threads in time slices, and by default the system timer ticks about every 15.6 milliseconds. Programs that sleep or wait for short intervals may actually wait longer than requested, in multiples of that granularity.
For trading software, it means a sleep of 1 millisecond might last 15 milliseconds. An EA using short sleeps for polling can behave less precisely than expected, and timing measurements can show odd quantization.
Some applications request a higher timer resolution, which reduces the interval to 1 millisecond at the cost of more power use. Others do not, and the behavior depends on whichever program on the system requested it.
For latency-sensitive setups, checking the current timer resolution is worthwhile. Utilities exist that display and adjust it. On a dedicated trading machine or VPS, a fixed high-resolution setting can improve timing consistency.
When analyzing timestamps and measuring latency, remember the quantization effect. Differences below the timer resolution may be artifacts.
Has the system's timer resolution ever been checked, and did it influence the latency data?
For trading software, it means a sleep of 1 millisecond might last 15 milliseconds. An EA using short sleeps for polling can behave less precisely than expected, and timing measurements can show odd quantization.
Some applications request a higher timer resolution, which reduces the interval to 1 millisecond at the cost of more power use. Others do not, and the behavior depends on whichever program on the system requested it.
For latency-sensitive setups, checking the current timer resolution is worthwhile. Utilities exist that display and adjust it. On a dedicated trading machine or VPS, a fixed high-resolution setting can improve timing consistency.
When analyzing timestamps and measuring latency, remember the quantization effect. Differences below the timer resolution may be artifacts.
Has the system's timer resolution ever been checked, and did it influence the latency data?