[ntar-workers] Timestamp resolution: if_tsaccur option unclear!

[Ulf Lamping]

Stephen Donnelly wrote:

>Depends what you want to measure?
>  
>
As always ...

>My opinion is that the minimum timestamp accuracy required for packet
>measurement is that minimum sized packets with minimum inter-packet gap
>should always receive unique tmiestamps, e.g. timestamps on sequential
>packets should always be strictly increasing.
>  
>
Well, having destinated hardware to measure times might take a different 
light ...

>For Gigabit Ethernet with minimum Ethernet frame size of 64 Bytes (plus
>IFG etc) this is approximately 576ns. This implies that the microsecond
>resolution of libpcap is already insufficient.
>  
>
Yes, I think that's right.

>For 10GE this would be approximately 58ns, implying that nanosecond
>resolution would be more than sufficient.
>
>I think the highest resolution that is likely to be useful is the
>byte-time of the link layer, since most link layers are essentially byte
>oriented (bitwise HDLC is an exception). In this case the resolution
>required for Gigabit Ethernet would be 8 nanoseconds, and for 10GE 1
>nanosecond.
>  
>
So the whole discussion get's esoteric for my purposes :-)

I just wanted to point out that there's a limit (probably not a problem 
for today). I just wanted to point out what the limits are and that they 
should be documented right today.

... you remember, the year 2000 problem is far, far away ;-)

Regards, ULFL