[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <d696a6da-022a-654a-d676-03e70df145aa@stressinduktion.org>
Date: Tue, 7 Jun 2016 13:23:21 +0200
From: Hannes Frederic Sowa <hannes@...essinduktion.org>
To: Krishna Mohan <krishnamohany3k@...il.com>, netdev@...r.kernel.org
Subject: Re: Fwd: Loopback properties in routing on linux kernel
On 07.06.2016 12:46, Krishna Mohan wrote:
> HI All,
> I have a question to determine the uses of multiple loopbacks
> interfaces(int loopback) in routing(quagga) on linux kernel. Please
> apologise if it is a dumb question. This is what I understand, please
> clarify if my understanding is correct.
>
> 1. we need them to represent a /32 ip, normally used for router-id.
> 2. we can measure traffic destined to that ip, via its independent
> statistics per loopback interface.
> 3. ip unnumbered uses loopbacks to represent that ip.
>
> my question is, is it necessary to have ifindex per loopback interface
> we create? if I create kernel loopback as lo:1 (as subinterface on
> default loopback in kernel) with ip assigned to it will do 1. for
> getting 2 and 3 done, loopback needs to be a separate netdev. is that
> correct?
That is correct, you can terminate the ip addresses for iBGP (or
router-id in this case) connections easily on any ip address that is
bound to some interface within the kernel. Note, that you don't
necessarily get the statistics correctly updated if you short-cut the
packet internally in the kernel. So the statistics on the dummy or
loopback device might not be accurate in such a setup.
> Will I be having any advantage if I configure the interface as lo:1 vs
> a separate netdev per loopback interface (i mean dummy netdev) to
> terminate the /32 destined traffic?
>From kernel PoV it doesn't matter. It is a question of taste and what
your tools prefer.
Bye,
Hannes
Powered by blists - more mailing lists