[Click] Re: click load balancer ?

[Maxwell Krohn]

Eddie Kohler (kohler at icir.org) wrote:
> 
> Yes! RoundRobinIPMapper does what you want. (It doesn't use a hash, though,
> it allocates connectoins round-robin.)  See our OPENARCH paper
> (www.icir.org/kohler/pubs/) on Click's general NAT support.
>

Hi Eddie, Thanks for your reply.  I've noticed a couple of things. 
First, the configuration for RoundRobinIPMapper seems to have changed
from the OPENARCH paper.  It used to be:

	rr_mapper :: RoundRobinIPMapper (
		pattern - - 10.0.0.8 80 1 0,
		pattern - - 10.0.0.9 80 1 0,
		pattern - - 10.0.0.10 80 1 0 );

but it seems like the most recent click release expects something along the
lines of:

	rr_mapper :: RoundRobinIPMapper (
		- - 10.0.0.8 80 1 0,
		- - 10.0.0.9 80 1 0,
		- - 10.0.0.10 80 1 0 );

(i.e., no "pattern" keyword).  Does this seems correct?

Also, as far as I can tell, the recent release never calls 
RoundRobinIPMapper::get_map because IPMapper::get_map is not
virtual.  I assume this should be virtual?

One last (higher level) question.  In effort to build this
load-sharing NAT configuration, I'm basing it on the Mazu-nat
skeleton, and substituting an rr_mapper for
the intern_server-related patterns.  Does this seem like 
a reasonable approach?

Many thanks / happy new year...

Max