Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 21 Apr 2011 15:23:46 -0700
From:      Garrett Cooper <yanegomi@gmail.com>
To:        Arnaud Lacombe <lacombar@gmail.com>
Cc:        Alexander Motin <mav@freebsd.org>, FreeBSD-Current <freebsd-current@freebsd.org>
Subject:   Re: Devices numbering [Was Re: Switch from legacy ata(4) to CAM-based ATA]
Message-ID:  <46AD49C8-FC7E-45FA-927D-5F1D912AB0BA@gmail.com>
In-Reply-To: <BANLkTimVp%2BpgMnbTt72vL0rhXRMjARfMzQ@mail.gmail.com>
References:  <BANLkTimVp%2BpgMnbTt72vL0rhXRMjARfMzQ@mail.gmail.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On Apr 21, 2011, at 10:48 AM, Arnaud Lacombe <lacombar@gmail.com> wrote:

> Hi,
>=20
> On Thu, Apr 21, 2011 at 1:31 PM, Garrett Cooper <yanegomi@gmail.com> wrote=
:
>> Although this may not be a list of fixable issues, here are some observat=
ions (in part with the new geom raid infrastructure):
>> 1. Channels are no longer fixed of course because ata uses cam now, and I=
 believe that device numbering is done based on probe ordering. This is fun t=
o work with when dealing with appliances or configurations that require dete=
rministic probe and mount, especially when drives fail, go missing, etc, but=
 can be hacked around in device.hints. This is why it would be nice for geom=
 labels to work in a sane manner.
> Out of context, but the same issue appear with network interfaces. If
> you're got 6 networks interface and the 3rd chip die, the 3 last get a
> bad numbering. I am not sure it is fixable by any device.hints.
>=20
> The Linux' world has the same issue (well, worse actually, as all
> interfaces uses the same 'eth' name). RedHat has been/will be
> introducing "Consistent Network Device Naming"[0,1] in Fedora 15,
> which may be an interesting move.

I'll have to look at this more in-depth, but this could be promising -- depe=
nding on how portable the biosdevhelper tool is.

Thanks,
-Garrett

PS while we may have a similar problem, at least the BSD interface names are=
 more sane than the Linux ones. Overall it would be nice to alias multiple l=
ogical names like "management", "data", etc, like we may present to customer=
s, but I don't know if someone has come up with an intelligent way to map ou=
t system topologies yet. It would be nice if serial numbers were exposed usi=
ng a vendor independent mechanism.=20=



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?46AD49C8-FC7E-45FA-927D-5F1D912AB0BA>