[ntp:questions] how will this effect stability

David L. Mills mills at udel.edu
Fri Feb 16 17:45:49 UTC 2007


Ry,

If you need the best time, traffic shaping is the worst thing to do. It 
introduces artificial delays in order to smooth the network flow. The 
clock filter and popcorn spike suppressor in the current NTP software 
design do in fact dramatically attenuate occasional delay spikes. 
However, for prolonged asymmetric congestion the huff-'n-puff filter can 
have a dramatic effect. It is much better than a crafted poll program 
that might attempt to find traffic holes.

Dave

Ry wrote:
> On Feb 14, 4:24 pm, Uwe Klein <uwe_klein_habertw... at t-online.de>
> wrote:
> 
> 
>>I am still thinking about patching ntpd to force polling at times
>>where i can guarantee an idle ( low delay ) line.
>>
> 
> 
> Wouldn't traffic shaping make more sense? We use the traffic shaping
> features on my company's firewalls and routers to boost NTP priority.
> NTP gets slightly higher priority than our VoIP phone system, which is
> several times higher than all other traffic (web, email, etc.). If a
> line is clogged (meaning router queues are full), NTP packets get
> first consideration.
> 
> The easiest-to-use trafic shaping systems simply modify the fair-
> queueing algorithms already present in a routing device... what sort
> of routers will you have? If they're Linux or BSD-based, you have a
> lot of options. If they are commercial products, consider making sure
> you get the QoS options. In my experience, traffic shaping is often
> required on a shared but low-bandwidth line.
> 
> Note that you may need traffic shaping at both ends of your slow
> connection for it to be effective, depending on your appliations. If
> you are connecting to an ISP directly, you might ask them what your
> options are.
> 
> Regards,
>   Ryan
> 




More information about the questions mailing list