Andy Gospodarek [off-list ref] wrote:

On Wed, Feb 28, 2007 at 02:39:42PM -0800, Jay Vosburgh wrote:

[...]

quoted

	Why would you want to turn this off?

When you connect active-backup bonds to 2 separate switches that are in
'distant' parts of the network you can end up with a bunch of unwanted
multicast data flowing everywhere and if you don't care whether or not
your multicast traffic is highly available then it just seems like
noise.  I thought the flexibility seemed nice.

	Ok, I can buy the "multicast spew" argument.

quoted

	Also, I've got a replacement patch for this functionality that
seems to be better in all regards. It sends bonus IGMP joins when a
failover occurs, rather than simply duplicating them on all slaves (the
current system can leave switches in the dark if the slaves fail back to
the originals).  As chance would have it, I'm planning to post it as
part of a set in a a little while.

That sounds like a nice add-on to the existing functionality.  I can see
the value in something dynamic like that, but I can also see the value
in something static like the functionality we have.  Did you plan to
keep the existing functionality intact or just have it done dynamically?

	Well, I posted the patch just a bit ago, so you can see for
yourself, but no, it removes the existing "copy IGMP everywhere"
behavior.  I couldn't really think of an advantage to flooding
everywhere all the time if the hose is re-aimed during failover (if
you'll pardon my cheesy metaphor).

Is this separate from your workqueue/refactoring patch or does it work
on the existing code?

	This is separate, for the current mainline.

	-J

---
	-Jay Vosburgh, IBM Linux Technology Center, fubar@us.ibm.com