[OpenIndiana-discuss] Same networking with varied NICs

[Michael Stapleton]

Hi Jim,

Sounds to me like DTrace is the tool for you.

http://docs.oracle.com/cd/E19253-01/817-6223/chp-intro/index.html

It's not something you learn in 5 minutes, but it really is worth the
effort.


IPMP does not have to be configured with test addresses. IPMP will uses
"Link based" by default if your NIC drivers support link state
notification.
The documentation really needs work... I think the documentation shows
examples of "Probe based" because it is much more difficult to configure
than "Link based".
Simply add the interfaces to the same IPMP group. Done.. 
But, IPMP is really about uptime, not band width. If you want to use
multiple interfaces, they all need IP addresses. They just do not have
to have "Test" addresses.

To use Aggregations, your switch and device drivers must support it.
I think your NIC drivers have to be GLDv3 compliant.


Mike


On Sat, 2012-10-20 at 15:38 +0400, Jim Klimov wrote:

> As I wrote earlier, I am trying to match-and-use varied
> drivers for my computer's NIC, one at a time, transparently
> to other users of the system (including vnics over this link).
> 
> FWIW, I tried to solve the problem differently than discussed
> before: now I tried to make an aggregation from instances of
> the different drivers. VNICs over the aggregation do seem to
> work, as well as "normal" communications from the GZ to internet,
> though the intermittent hangs do still appear with both stock
> rge and gani drivers for the builtin Realtek 8168/8111 GbE NIC.
> 
> The problem is that the aggregation can only be defined over
> existing interfaces (also if directly hacking into the config
> file /etc/dladm/datalink.conf), so basically I can't predefine
> an "aggr0" over "rge0 + gani0 + e1000g0" and have it work with
> whatever driver I currently have loaded. If I mention a driver
> which is not present at the moment (i.e. "rge0" while "gani" is
> loaded, or "e1000g0" while in physical hardware boot) the aggr0
> link is not spawned at all.
> 
> I looked at IPMP, but it seems too unwieldy for the laptop case
> (each component of an ipmp group needs its own IP and a known
> external node to test against).
> 
> Can the aggregations be forced to accept missing devices and
> work with those currently available? If that's not currently
> possible, does it seem like a good RFE (i.e. hardware can break
> so upon a reboot a server's NIC really can go missing - would
> be bad to lose a whole aggr because of that)?
> 
> Also, config changes made with dladm program take their place
> immediately, however hacks into its config file require a reboot.
> How can I make the system re-read the /etc/dladm/datalink.conf
> file and apply manually changed settings? I tried to run
> "svcadm restart datalink-management" - did not help...
> 
> 2012-10-01 14:02, Jim Klimov wrote:
> > Hello all,
> >
> >    I wondered what is a Crossbow+VanityNaming way of
> > doing some things I've done some time ago with static
> > config files:
> >
> >    I have an installed OS image which can be booted on
> > different hardware (say, a pass-through partition with
> > OI that can be booted from hardware BIOS as a dual-boot
> > option, as well as in a VM from another host OS on the
> > box). In these different hardware environments this box
> > sees varied networking gear - an rge0 in one case and
> > an e1000g0 in another. I want the logical networking
> > to be the same in these cases.
> >
> >    So, previously for a singular global zone I made two
> > files (/etc/hostname.e1000g0 and /etc/hostname.rge0)
> > with identical contents, and the system plumbed the
> > one "present" NIC with the needed addressing setup.
> >
> >    Now I want to make some VNICs and configure some
> > local zones with attachment to the external NIC (can
> > play with both exclusive and shared IP stacks).
> >
> >    I expected that "vanity naming" can help me in this
> > case by naming "the present NIC" for example "eth0",
> > and my zone and VNIC attachments would go over eth0.
> > Should this work?
> >
> >    Alternately, can I create an etherstub with several
> > attached VNICs, including one with IP configuration
> > for the global zone, and bridge it to external LAN
> > via "the one present NIC" (i.e. by attaching both
> > e1000g0 and rge0 instances to the etherstub)?
> >
> >    Disclaimer: I did not yet try either variant, and
> > am in the process of setting up the boot of physical
> > OI from a VM in another OS, but wanted to know in
> > advance what to expect ;)
> >
> > Thanks,
> > //Jim
> 
> 
> _______________________________________________
> OpenIndiana-discuss mailing list
> OpenIndiana-discuss at openindiana.org
> http://openindiana.org/mailman/listinfo/openindiana-discuss