[ntp:hackers] Fw: Internet delays affecting NTP timing - feature or bug?
Greg.Dowd at microsemi.com
Tue May 28 19:56:32 UTC 2019
I don't know how deep you have researched this but there are a couple simple things which can impact ntp. First, I assume you are not using dns names for the server which can bounce, rotate or change. Are these clients using global addresses? If changing the ip address changes the behavior, I would suspect a load balancer somewhere in the path. They typically use a hash of source ip and port to determine which port to forward through. This can result in a significant different in rtd and asymmetry. If you are NAT'd, your source ports may be getting remapped. A lot of little things that impact path, delay and asymmetry and typically aren't noticeable until you start studying timestamps.
From: hackers [mailto:hackers-bounces+greg.dowd=microsemi.com at lists.ntp.org] On Behalf Of Danny Mayer
Sent: Tuesday, May 28, 2019 7:57 AM
To: Peter Martinez <peter.martinez at btinternet.com>; hackers at lists.ntp.org
Subject: Re: [ntp:hackers] Fw: Internet delays affecting NTP timing - feature or bug?
Can you send details of this to at least Miroslav and myself so we can look at what you are seeing? It's quite possible that the networks are not giving the NTP packet priority as it isn't coming from the NTP port.
Pure speculation on my part right now without knowing any details.
On 5/26/19 5:46 AM, Peter Martinez wrote:
> Repeating my last message:
> ----- Original Message ----- From: "Peter Martinez"
> <peter.martinez at btinternet.com>
> To: <hackers at lists.ntp.org>
> Sent: Thursday, May 23, 2019 7:14 AM
> Subject: Internet delays affecting NTP timing - feature or bug?
>> Greeting to the Hackers list from a new member.
>> I have recently been writing code for NTP, both client and server, as
>> a follow-up to building a GPS-based time reference. I wanted to use
>> the GPS to explore the performance of the NTP system globally. I
>> worked from the RFCs and have not studied the "reference" software.
>> I have noticed a strange and worrying phenomenon which I would like
>> to describe and see if others can see it too. Maybe something needs
>> to be done about it.
>> I first spotted that if I poll a distant NTP server from an NTP
>> client on one PC on my LAN, I see a different offset and delay
>> compared with another instance of the same client, either on the same
>> PC or another PC on the same LAN. But ONLY on a small number of
>> distant servers. This led to the discovery that it was difference in
>> the local port addresses of the two PCs that was triggering the
>> effect - by modifying the client software so that I could change the
>> local port address, I discovered that polling one of the affected
>> servers then changing the port address, would change the indicated
>> offset and delay. On other servers, changing the port address has no
>> I have a lot more detail I can describe, but for this post I just
>> want to ask if anyone else has seen this. If not, maybe someone
>> would care to look for it. Modify your NTP client software to let
>> you choose the local port address in real time, poll a few servers
>> around the world and see what shows up.
>> I am fairly sure that this isn't a bug in my software, and I am
>> fairly sure it's not in the affected servers. It seems to be "out
>> there" in the big wide internet. If I am right, then this will
>> affect NTP accuracy. I could ask the question "how is it that some
>> part of an internet route can introduce a spurious delay to a packet
>> of data which is triggered by the content of the packet?". Is it a
>> feature or a bug? If it's a bug, can it be fixed? NTP will be better
>> if it can be fixed.
>> Peter Martinez
>> NW England
> hackers mailing list
> hackers at lists.ntp.org
hackers mailing list
hackers at lists.ntp.org
More information about the hackers