[Babel-users] RTT stability inside a GRE tunnel

[Baptiste Jonglez]

On Tue, Jul 02, 2013 at 09:35:55AM -0400, Michael Richardson wrote:
> 
> Can you tell us a bit how about how you are measuring, and what the remote
> side is?   I see that the two values are consistent with each other --- it
> looks to me like the inside tunnel value is simply truncated.
> 
> I assume that you are measuring the RTT value from babeld, rather than from
> the network itself (i.e. with tcpdump).

Both measures are done via ICMP; in each case, it is a simple,
continuous ping with a 1-second interval.

The hosts are running Debian stable; they are both behind a NAT, and
the GRE tunnel is setup over IPv4 (which the firewall/NAT on front of
each host seems to let through without any configuration).

It would be interesting to compare with the RTT measured with babel,
but it only works through the tunnel (IPv6 multicast, which babeld
uses to talk with its neighbours, isn't routed on the Internet).

> This is what I would *suspect*:
> 
> Coming out of the GRE tunnel, your packet would have to be decapsulated by
> the kernel, and then re-queued into the network bottom half.  Generally,
> the kernel does interrupt mitigation, and since this is a soft-interrupt,
> that queue is serviced from a clock interrupt.
> 
> As such, you'd see the values quantized to the clock interrupt, which is
> exactly what I see in your diagram.

As Denis pointed out, this is unlikely, but thanks for pointing out
this interesting idea.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: not available
URL: <http://lists.alioth.debian.org/pipermail/babel-users/attachments/20130702/f28a50de/attachment.sig>