I am working on some applications that route tcp and update traffic across several machines, and I am trying to timestamp (millisecond resolution) packets on each node to get a complete picture of the delay through each node.
Before the tests, I made sure that all Windows machines are synchronized using the NTP protocol, but my problem is that the time on all machines is never fully synchronized, and the system times seem to change +500 ms. For example, sometimes my timestamp logs show that a packet was received at node 2500 ms before it was sent from node1.
After re-synchronizing the NTP of all machines, different latencies between nodes1 and node2 will always be displayed, but the time differences are always constant until the next NTP synchronization.
What accuracy can be expected when synchronizing multiple computers with the same NTP server? Is there a reason why I see these discrepancies and how other applications measure latency through a system that includes multiple machines?
Thank,
Tom
source
share