Packet loss on peer associations? - NTP

This is a discussion on Packet loss on peer associations? - NTP ; All, I have recently started peering two of my NTP servers. One is a primary server, the other only has Internet-based sources of synchronisation. They are attached to the same LAN, which is lightly loaded, as are the servers themselves ...

+ Reply to Thread
Results 1 to 3 of 3

Thread: Packet loss on peer associations?

  1. Packet loss on peer associations?

    All,

    I have recently started peering two of my NTP servers. One is a primary
    server, the other only has Internet-based sources of synchronisation.
    They are attached to the same LAN, which is lightly loaded, as are the
    servers themselves (CPU usage rarely exceeds 1% and there is little disk
    activity).

    Quite often though, the association misses a beat (e.g. the reach number
    reported by ntpq -p is 376 or another combination of mostly 1s and few
    0s). I would expect to see this in the presence of packet loss, but I
    really don't think that that is the case here.

    Note that I have clamped both maxpoll and minpoll to 6 on both sides
    (because otherwise the problem was even worse, whenever the two peers
    did not agree on the poll interval).

    Is this a problem?

    (One server is running ntpq 4.2.3p41@1.1401-o under Linux 2.4.20 on a
    P4, the other ntpq 4.2.2p4@1.1585-o under Linux 2.4.31 on a soekris Geode).

    Thanks, Jan

  2. Re: Packet loss on peer associations?

    > All,
    >
    > I have recently started peering two of my NTP servers. One is a primary
    > server, the other only has Internet-based sources of synchronisation. They
    > are attached to the same LAN, which is lightly loaded, as are the servers
    > themselves (CPU usage rarely exceeds 1% and there is little disk
    > activity).
    >
    > Quite often though, the association misses a beat (e.g. the reach number
    > reported by ntpq -p is 376 or another combination of mostly 1s and few
    > 0s). I would expect to see this in the presence of packet loss, but I
    > really don't think that that is the case here.
    >
    > Note that I have clamped both maxpoll and minpoll to 6 on both sides
    > (because otherwise the problem was even worse, whenever the two peers did
    > not agree on the poll interval).
    >
    > Is this a problem?
    >
    > (One server is running ntpq 4.2.3p41@1.1401-o under Linux 2.4.20 on a P4,
    > the other ntpq 4.2.2p4@1.1585-o under Linux 2.4.31 on a soekris Geode).
    >
    > Thanks, Jan


    I have the same question with a similar setup. I have two Stratum 2s peered
    with each other. Both are Linux machines running on the same LAN with
    little traffic on the LAN itself, both have light system loads, both are
    running ntp 4.2.4, and each have their own sychronization sources. One
    server is a pool server and the other server is not visible to the internet.
    Running 'ntpq -p' on each server routinely shows its peer with a reach of
    376, while the rest of the synchronization sources show 377. I also don't
    believe packet loss is an issue here. Is this typical behavior with peered
    servers?

    Thanks,

    Dennis



  3. Re: Packet loss on peer associations?

    Jan,

    It is normal that one or the other peers in symmetric modes appear to
    lose a packet. Not so; it might happen that the poll intervals don't
    completely match and one peer occasiionally transmits two packets while
    the other one transmits one. The protocol exchange is specially designed
    to detect what appear to be duplicate, missing or out of order packets.

    If the poll intervals happen to be different, the protocol selects the
    minimum value of the two. If you leave things alone and the intervals
    finally climb to 1024 s while mismatching along the way, no harm is done.

    Dave

    Jan Ceuleers wrote:

    > All,
    >
    > I have recently started peering two of my NTP servers. One is a primary
    > server, the other only has Internet-based sources of synchronisation.
    > They are attached to the same LAN, which is lightly loaded, as are the
    > servers themselves (CPU usage rarely exceeds 1% and there is little disk
    > activity).
    >
    > Quite often though, the association misses a beat (e.g. the reach number
    > reported by ntpq -p is 376 or another combination of mostly 1s and few
    > 0s). I would expect to see this in the presence of packet loss, but I
    > really don't think that that is the case here.
    >
    > Note that I have clamped both maxpoll and minpoll to 6 on both sides
    > (because otherwise the problem was even worse, whenever the two peers
    > did not agree on the poll interval).
    >
    > Is this a problem?
    >
    > (One server is running ntpq 4.2.3p41@1.1401-o under Linux 2.4.20 on a
    > P4, the other ntpq 4.2.2p4@1.1585-o under Linux 2.4.31 on a soekris Geode).
    >
    > Thanks, Jan


+ Reply to Thread