Skip to content

CONCEPT Cited by 1 source

Amazon Time Sync attribution

Amazon Time Sync is AWS's fleet-wide time-synchronisation service, exposed as a local NTP/PTP endpoint on Nitro-based EC2 instances. It provides sub-millisecond accuracy across the fleet, making wall-clock time ranges a reliable attribution key.

Why this matters for attribution

Heartbeat-based ownership represents the owner of a resource as a list of (owner, t_start, t_end) time ranges, and attribution is done by looking up the query's timestamp against the range list. This design depends on the following being true:

  • The timestamp on an observation (the flow's t_start / t_end) was taken on a clock that agrees with the clock used to stamp heartbeats to within the granularity needed.
  • Two hosts reporting the same resource's ownership can agree on the ordering of transition events.

If host-level clocks drifted by even hundreds of milliseconds, the time-range lookup would straddle boundaries or miss the correct range — misattributions and unattributions would multiply. Amazon Time Sync's sub-ms accuracy makes the clock a non-issue for AWS-hosted fleets.

"Since Netflix uses Amazon Time Sync across its fleet, the timestamps (captured by FlowExporter) are reliable."

Boundary

  • Across AWS regions. Amazon Time Sync is regional; cross-region traffic relies on the fact that all regions ultimately reference the same atomic-clock source, but clock-comparison-across-regions has higher uncertainty. Netflix's cross-regional attribution path uses forwarding + regional time-range lookup, so the lookup happens in the target region's clock frame.
  • Across cloud providers. Not a concern for AWS-native Netflix.

Seen in

Last updated · 319 distilled / 1,201 read