Subject: Re: IP Multipathing (IPMP) For NetBSD
To: Hubert Feyrer <hubert@feyrer.de>
From: matthew sporleder <msporleder@gmail.com>
List: tech-net
Date: 06/07/2007 09:23:32
On 6/7/07, Hubert Feyrer <hubert@feyrer.de> wrote:
> On Thu, 7 Jun 2007, Darren Reed wrote:
> > Is there anyone out there interested in implementing a solution for
> > NetBSD that implements IPMP ?
> >
> > A very brief description of IPMP can be found at:
> > http://en.wikipedia.org/wiki/IPMP
>
> As far as I understand from that page, it has two parts:
> #1 failover if a network card explodes
> #2 link aggregation.
>
> If you just want #2, check out agr(4), which will be in 4.0.


(I'm very interested in this, as I've asked this question to netbsd
mailing lists a few times over the years)

IPMP provides, in its most basic configuration, a (very seamless)
failover of an IP to another nic in a named "group" if the link state
changes on the "active" nic.  (i.e. if the cable is pulled)
(link-only ipmp is new in solaris 10, as far as I can tell)

It can then be configured to failover with link state AND provide link
aggregation in a named group, but this requires that you setup probes
to test various gateways.  (on solaris, this part is extremely prone
to human error because not having enough probe targets can cause
unintended outages when various environmental things happen, netbsd
could improve on this)

You can also use probe-based testing without aggregation.

Personally, I think the failover part of IPMP is the most important.
We could also do this better than the linux bond-ing system (basically
the same thing), which I've always thought was a little kludgey.

_Matt