Mailing List Archive: load balancing over a very large number of outgoing addresses?

Hi,
I have recently developed a need to multiplex connections from within a
NAT over several (hundred, even thousand if possible) external IPs. I
can have all of these IPs routed to a single interface on my NAT box,
however I am not exactly sure how to set up a random/round robin load
balancing scheme such that outgoing connections from my network each get
a random source address from my source address pool.

I have come across this how-to:
http://lartc.org/howto/lartc.rpdb.multiple-links.html

However it seems that I cannot scale these routing rules past 255
routes, and unlike the example, I am not multiplexing interfaces but
only IPs in roughly a contiguous /16 range being routed to this linux
machine.

Any suggestions on how to get this up and running would be greatly
appreciated.

Many thanks,
Chris Kanich

On 07/26/07 20:27, Chris Kanich wrote:
> I have recently developed a need to multiplex connections from within a
> NAT over several (hundred, even thousand if possible) external IPs. I
> can have all of these IPs routed to a single interface on my NAT box,
> however I am not exactly sure how to set up a random/round robin load
> balancing scheme such that outgoing connections from my network each get
> a random source address from my source address pool.

If I understand what you are wanting to do correctly, that is many to
many NAT, why not use a range of IP addresses on your SNAT rule? I.e.:

iptables -t nat -A POSTROUTING -j SNAT --to-source A.B.0.1-A.B.7.255

Would SNAT to an IP in the range of A.B.0.1 through A.B.7.255, thus a
little over 2000 IPs.

One thing I'm not sure of is how the kernel decides which IP in the
range to assign, though I bet someone on this mailing list can help
better answer this.

> However it seems that I cannot scale these routing rules past 255
> routes, and unlike the example, I am not multiplexing interfaces but
> only IPs in roughly a contiguous /16 range being routed to this linux
> machine.

*nod* I don't think this is what you are wanting to do.

> Any suggestions on how to get this up and running would be greatly
> appreciated.

See if what I presented above is any where close to what you are wanting
to do.

Grant. . . .