Lucky Packets and Network Time Protocol (NTP)

Published

January 26, 2026

Author(s)

Jace Smuszkiewicz, Terrence Jones, Andrew Novick, Demetrios Matsakin

Abstract

For users of the network time protocol (NTP), a "lucky packet" is defined as one which is returned back to the client with very short roundtrip travel time over the internet. In NTP the error budget of the time offset measurement is dominated by the asymmetry in the network path, and it can be shown that its associated error must always be less than half the roundtrip time [Mills, 2011]. Therefore it is often stated that it is best to use only the luckiest paths, and plots indicating the improved precision from retaining only lucky packets are often shown. Although the shortest path is not necessarily the most symmetric, precision improvements support the idea that it is good to retain only the luckier packets. In an effort to determine what level of lucky packets it is best to retain, or highly weight, we queried timing labs in the Americas, Europe, and Asia from NTP clients in Boulder, Jamaica, Saint. Louis, and Washington D.C., realizing that in operations queries should not be made too frequently. By querying servers directly referenced to either a UTC realization or GNSS, we were able to study the relation between measured time offset and roundtrip time. We often do not see the expected pattern in which measured NTP time offsets would be distributed over the range of plus or minus half the roundtrip time (creating a filled arrow-head shape when the measured time offset is plotted against roundtrip time). Rather, the pattern often showed linear and sometimes banded relationships between roundtrip time and timing offset. The likely possibility that the observed timing variations are due to semi-systematic internet routing variations is explored by tracing the packet paths with standard Windows and Linux tools. Although the minimum roundtrip time was rarely associated with near-zero time offset, as a general strategy and in the absence of server-specific information the best practice may be to query only nearby servers from national timing labs and retain only the luckiest 10 % of the data if one is doing simple daily averaging in post-processed mode, as would be useful for verifying traceability. Alternately a running average or Kalman filter could weigh new data with a function that assigns decreased weight to points with roundtrip times outside the tenth percentile of a previously-determined distribution. A different weighting can be used if the user is confident of the roundtrip/assymetry pattern for any server-client pair.

Proceedings Title

ION PTTI 2026 Proceedings - 57th Annual Precise Time and Time Interval Systems and Applications Meeting

Conference Dates

January 27-29, 2026

Conference Location

Anaheim, CA, US

Conference Title

PTTI Systems and Applications Meeting

Pub Type

Conferences

Download Paper

Local Download

Keywords

network time ptotocol, time transfer

Citation

Smuszkiewicz, J. , Jones, T. , Novick, A. and Matsakin, D. (2026), Lucky Packets and Network Time Protocol (NTP), ION PTTI 2026 Proceedings - 57th Annual Precise Time and Time Interval Systems and Applications Meeting, Anaheim, CA, US, [online], https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=961509 (Accessed February 26, 2026)