Multi-Interface Patch

dormando dormando at
Wed Feb 13 05:07:29 UTC 2008

Dustin Sallings wrote:
> On Feb 12, 2008, at 18:38, dormando wrote:
>> If any of these are worth addressing, let me know. Otherwise it's
>> going in with some minor style fixes for 1.2.5.
>     We should probably have a plan for this sort of thing for the binary
> branch.  This patch makes the tree very sad as it is.

Sorry; I was going to start blowing time getting the binary tree up to
date again. Or develop it separately for a while until we get rid of
performance regressions. Getting it up to date is totally cool either way :P

>     Should we assume that every interface that binds for the text
> protocol should also bind for the binary protocol?  That seems pretty
> straightforward, and will probably be consistent with the UDP strategy.

I guess so? Little lost in the code here.

>     I've not been paying much attention to this thread, honestly.  Is
> there a short summary of the goal of explicit multiple interface
> binding?  It doesn't seem like it does much over IN_ADDR_ANY in the
> general case but become a multiplier for file descriptors (once there's
> a binary UDP implementation, it'd be n interfaces * m layer 3 protocols
> * 4).
>     I'm not so much questioning its value as I am curious.  Right now,
> it just looks like a lot of work for me.  :)

It's mainly for ipv6 support... "Proper" binding for listening sockets
means you bind to every socket that the address lookup returns.
Presently we just bind on the first one that returns, which is usually
fine for ipv4. Apparently it's a bigger deal on ipv6.

This is also to help start plumbing for ... any other measure of
multiple interface support.


More information about the memcached mailing list