Re: [RFC PATCH net-next] bonding: add a vlan+srcmac tx hashing option

From: Jarod Wilson <jarod@redhat.com>
To: Jiri Pirko <jiri@resnulli.us>
Cc: linux-kernel@vger.kernel.org, Jay Vosburgh <j.vosburgh@gmail.com>,
	Veaceslav Falico <vfalico@gmail.com>,
	Andy Gospodarek <andy@greyhouse.net>,
	"David S. Miller" <davem@davemloft.net>,
	Jakub Kicinski <kuba@kernel.org>, Thomas Davis <tadavis@lbl.gov>,
	netdev@vger.kernel.org
Subject: Re: [RFC PATCH net-next] bonding: add a vlan+srcmac tx hashing option
Date: Fri, 8 Jan 2021 10:21:17 -0500	[thread overview]
Message-ID: <20210108152117.GC63172@redhat.com> (raw)
In-Reply-To: <20210108131256.GG3565223@nanopsycho.orion>

On Fri, Jan 08, 2021 at 02:12:56PM +0100, Jiri Pirko wrote:
> Fri, Jan 08, 2021 at 12:58:13AM CET, jarod@redhat.com wrote:
> >On Mon, Dec 28, 2020 at 11:11:45AM +0100, Jiri Pirko wrote:
> >> Fri, Dec 18, 2020 at 08:30:33PM CET, jarod@redhat.com wrote:
> >> >This comes from an end-user request, where they're running multiple VMs on
> >> >hosts with bonded interfaces connected to some interest switch topologies,
> >> >where 802.3ad isn't an option. They're currently running a proprietary
> >> >solution that effectively achieves load-balancing of VMs and bandwidth
> >> >utilization improvements with a similar form of transmission algorithm.
> >> >
> >> >Basically, each VM has it's own vlan, so it always sends its traffic out
> >> >the same interface, unless that interface fails. Traffic gets split
> >> >between the interfaces, maintaining a consistent path, with failover still
> >> >available if an interface goes down.
> >> >
> >> >This has been rudimetarily tested to provide similar results, suitable for
> >> >them to use to move off their current proprietary solution.
> >> >
> >> >Still on the TODO list, if these even looks sane to begin with, is
> >> >fleshing out Documentation/networking/bonding.rst.
> >> 
> >> Jarod, did you consider using team driver instead ? :)
> >
> >That's actually one of the things that was suggested, since team I believe
> >already has support for this, but the user really wants to use bonding.
> >We're finding that a lot of users really still prefer bonding over team.
> 
> Do you know the reason, other than "nostalgia"?

I've heard a few different reasons that come to mind:

1) nostalgia is definitely one -- "we know bonding here"
2) support -- "the things I'm running say I need bonding to properly
support failover in their environment". How accurate this is, I don't
actually know.
3) monitoring -- "my monitoring solution knows about bonding, but not
about team". This is probably easily fixed, but may or may not be in the
user's direct control.
4) footprint -- "bonding does the job w/o team's userspace footprint".
I think this one is kind of hard for team to do anything about, bonding
really does have a smaller userspace footprint, which is a plus for
embedded type applications and high-security environments looking to keep
things as minimal as possible.

I think I've heard a few "we tried team years ago and it didn't work" as
well, which of course is ridiculous as a reason not to try something again,
since a lot can change in a few years in this world.

-- 
Jarod Wilson
jarod@redhat.com