More desk clutter... :-)

[Gary E. Miller]

Yo Hal!

On Thu, 23 Mar 2017 14:24:25 -0700
Hal Murray <hmurray at megapathdsl.net> wrote:

> gem at rellim.com said:
> > Which, paradxically, is why GigE is usually better than 100Base-T.
> > At the same total amount of data per second the latency could be
> > cut by a factor of ten.  
> 
> > And less latency would hopefully lead to less jitter.   
> 
> Reducing the latency has no direct impact on the jitter.

I disagree.  Much of the jitter on a lightly loaded ethernet is
determined by message length(s).  In gpsd land this is easily seen that
faster serial speed reduces the jitter on NMEA time stamps.

Not totally applicaable here as NMEA time stamps are variable lenght
and NTP ethernet packets are fixed length.  But the other traffic
is getting in the way.

Of course there are also many unrelated sources of jitter as well.

And my practical experience has usally been that faster links, with the
same load, have less jitter.  The same load, at a faster speed, leads
to shorter Queues in the time domain.  Queue length variation leads to
jitter.

> There is a second order effect that depends on the network loading.
> If there is other traffic on the parts of the network used by your
> NTP traffic, the queuing delays will be shorter with a faster
> network.  On a lightly loaded network, the difference will be hard to
> measure.

Yes, the jitter reduction is partly load dependent.  I would call it more
a first order effect.

> There is a much bigger potential problem with gigabit networks.  In
> the early days, the per-packet interrupt overhead was excessive so
> they batched interrupts.  That can cause delays and jitter.

Yes, the devil is in the details.

> Do we have a clear description (or link) for how to disable that?

Disable what?  Batched interrupts?  I find that tuning ethernet ports is
very hardware and driver specific.

One thing that I have found often helps is turning off EEE.  If your
adapter or switch supports it, turn it off now.

Another ihelp s never polling any chimer slower than 60 seconds to keep
the arp and routing caches full.

I just looked at one of my ethernet ports, it has 50 different binary
options. Another has 44.  Then there are all the tunebles.  I could
get lost in there for a month...

RGDS
GARY
---------------------------------------------------------------------------
Gary E. Miller Rellim 109 NW Wilmington Ave., Suite E, Bend, OR 97703
	gem at rellim.com  Tel:+1 541 382 8588

	    Veritas liberabit vos. -- Quid est veritas?
    "If you can’t measure it, you can’t improve it." - Lord Kelvin
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 488 bytes
Desc: OpenPGP digital signature
URL: <https://lists.ntpsec.org/pipermail/devel/attachments/20170323/871e94b4/attachment.bin>