[time-nuts] SLIP vs Ethernet for NTP
Doug Calvert
dfc-list at douglasfcalvert.net
Mon Oct 24 05:16:54 UTC 2011
On 10/24/2011 12:26 AM, Magnus Danielson wrote:
> On 10/24/2011 05:14 AM, Doug Calvert wrote:
>> On 10/23/2011 08:37 AM, Tijd Dingen wrote:
>>> I seriously doubt that piping it through slip is going to help. You
>>> still get to deal with slip + tcp protocol stack. I would expect gbit
>>> ethernet with possibly some fifo/buffering setting adjusted gets you
>>> better performance that RS232 (with or without slip). Just assemble
>>> some raw frames on the source side and go.
>>>
>> I think gig ethernet has a higher delay for smaller packets (ntp's udp
>> datagram is tiny) than 100mbit ethernet.
>
> There is very little to support this claim. The line-encoding could
> possibly cause a small difference, but other than that I suspect
> implementation issues rather than protocol by itself.
>
> Even if you said that gigabit ethernet cards tend to have a higher delay
> for smaller packets I would react, but then not on the technical aspect
> but rather that I would like to see some statistics supporting the claim.
>
> Only real technical issue is that datagram size needs to be 64 bytes and
> that the header, trailer and inter-packet gap has the same size, but
> behaves like larger relative parts than for larger packets. However,
> this effect would be valid for fast ethernet too.
>
I am off to bed so i do not have time to track down some of the links. I
too was skeptical the first time I heard this. It has something to do
with the framing (among others) if memory serves me. It was recently
discussed on ntp-questions or ntp-hackers. The one link I do remember is
from mysql-clusters:
http://lists.mysql.com/cluster/3976
More information about the time-nuts
mailing list