[Server-devel] Trying to Understand the Why of Bonding in XS 0.5

[Martin Langhoff]

2008/12/8 Anna <aschoolf at gmail.com>:
> What I've gleaned so far - the purpose of bonding as far as the XS is
> concerned is to allow the APs attached to eth1 and the usb active antennas
> to work with dhcp in an environment where any of these devices may be
> plugged in at any point without restarting the dhcpd service or the
> network.  Correct?

Yes, but not just that.

 - We run DHCPd on 4 interfaces that may or may not be there. This
includes the 3 possible AAs and the eth1 NIC -- a server with AAs may
not have eth1

 - We run many services that attach to a notional "main ip address of
the XS". Apache, idmgr, backup service. That ip address must be on an
interface that is not tied to a particular hw.

And at the same time, we want a single configuration that just works
with whatever is there. The whole set of services comes up when the
server starts, regardless of hw. An AA appears? great, link it up to
the appropriate bonding device. This XS never had an "eth1" NIC, but a
technician came around and plugged on in? Fine, as soon as the machine
boots, eth1 "falls into place". Remove it? Just works again.

Way back in time we consulted with David Woodhouse and other core
linux devs - they suggested we used bridges like OpenWRT. You've seen
the discussion of their downsides and why we switched to bonding.

If bonding is really broken I'll find an alternative technique. But see below...

> Since Birmingham is planning on using "legacy equipment" for the XSs (read -
> "old stuff"), XS software compatibility with older hardware is key.

Legacy kit is high on my list too - many (most?) pilots run on it.
Still... bonding is a very well established part of the networking
stack in the linux kernel, I doubt there's a fundamental problem with
it - now that you've narrowed the problem down to it, let's see if we
can find a way to get more specific debug information.


> I don't mind trying to figure out stuff, but it sure is nice when things
> "just work," as networking mostly did in XS 0.4.

Thanks! I also yearn for things that Just Work. Let's see if we can
learn a bit more about this issue.

cheers,



m
-- 
 martin.langhoff at gmail.com
 martin at laptop.org -- School Server Architect
 - ask interesting questions
 - don't get distracted with shiny stuff  - working code first
 - http://wiki.laptop.org/go/User:Martinlanghoff