Re: libpcap timestamp for sending a packet ?

["Aaron Turner" <synfinatic () gmail com>]

Hate to follow up, but realized a mistake... NIC's with TCP offload
engines in hardware may put the TCP timestamp option in the header.  
I know from a co-worker that the Nvida TOE chipset does for example.

On 3/1/06, Aaron Turner <synfinatic () gmail com> wrote:
> No, NIC's don't put timestamps in packets.  And depending on your
> OS/NIC driver, the libpcap timestamp of the packet could be wrong or
> at least non-sensical too (I've seen packets stamped with an earlier
> timestamp then the previous read packet).
>
> IMHO, your best bet is to either use the TCP timestamp option on a
> socket() or use libnet to generate raw packets and put your timestamp
> in the payload.
>
> Btw, you do know that there are already tools which do this right???
> things like iperf and netperf.
>
> --
> Aaron Turner
> http://synfin.net/
>
>
> On 3/1/06, J S <geekreader () gmail com> wrote:
> > Hello
> > I am setting up a framework for measuring one way delay between two Internet
> > end points.
> > For higher accuracy I am using libpcap time stamps (from the packet header)
> > at the reciever.
> > Does any one know if there is any method for improving accuracy at the
> > sender side.
> > Is it possible for sender's NIC to log the time stamp the packet was sent?
> > or any other idea ?


--
Aaron Turner
http://synfin.net/
-
This is the tcpdump-workers list.
Visit https://lists.sandelman.ca/ to unsubscribe.