Re: [PATCH net-next 2/2] selftest: net: add icmp reply address test

From: Francesco Ruggeri <fruggeri@arista.com>
To: David Ahern <dsahern@gmail.com>
Cc: David Miller <davem@davemloft.net>,
	shuah@kernel.org, netdev <netdev@vger.kernel.org>,
	linux-kselftest@vger.kernel.org
Subject: Re: [PATCH net-next 2/2] selftest: net: add icmp reply address test
Date: Sat, 2 Nov 2019 15:08:19 -0700	[thread overview]
Message-ID: <CA+HUmGhYzSE-ruiOfQa9UCKcMuN361asxwCD=Nmdjar9jC0bTA@mail.gmail.com> (raw)
In-Reply-To: <CA+HUmGjra-=GeRApvYRgX6iQZPG73xWyfXqR-_fxjKS0WcmYrQ@mail.gmail.com>

> > > I apologize in advance  for being slow ...
> > > I have 3 namespaces that have to share the same LAN, I am not trying
> > > 1-1 connections among those namespaces.
> > >
> >
> > How would you cable this if it were an actual network with physical nodes?
> > - bridge on R1 (since it is the gw for H1), with connections to R2 and
> > H1 into the bridge
> > - second connection between R1 and R2
> > - connection between R2 and H2
> >
> > For the simulation, network namespaces represent physical nodes, veth
> > pairs act like a cable between the nodes / namespaces and the bridge
> > makes the LAN.

Thanks, I see what you mean now.
I was assuming a different physical model, with all the namespaces on the LAN
connected to a hub (simulated by the dummy device). For simulation purposes this
model seem simpler: there are N interfaces instead of N pairs, and one does not
have to deal with the bridge end of the pairs.
Why is the model you described preferable?

Thanks,
Francesco