LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: problems getting direct routing to work

To: Donald Ball <balld@xxxxxxxxxxxxxxx>
Subject: Re: problems getting direct routing to work
Cc: lvs-users@xxxxxxxxxxxxxxxxxxxxxx
From: Horms <horms@xxxxxxxxxxxx>
Date: Sun, 9 Apr 2000 20:27:05 -0700
On Sun, Apr 09, 2000 at 11:19:09PM -0400, Donald Ball wrote:
> http://linuxvirtualserver.org/arp.html
> 
> I added these rules to the ipchains configuration on the real server:
> 
> ipchains -A input -j REDIRECT -d 206.66.49.220 -p tcp
> ipchains -A input -j REDIRECT -d 206.66.49.220 -p udp
> 
> (206.66.49.220 is the virtual ip address, of course)

I've seen that idea somewhere before :)

You may want to look at using aliases on the loopack device
instead if you are using a 2.2.14 kernel as according to
Rusty and some anecdotal evidence there are performance issues
with having trandparent proxying compiled into the kernel, let
alone actually using the feature.

You can have a single ip alias cover a network using something along
the lines of

ifconfig lo:0 192.168.0.1 netmask 255.255.255.0 mtu 150

> and everything started working fine. It's a tad onerous to create the
> rulesets for all of the virtual ip addresses, but much simpler than
> maintaining 'n' seperate copies of the apache configuration file, so I'm
> happy.

If you look at the fwmark functionality in 0.9.10, and your
virtual ipaddresses are (somewhat) contiguous then you shoudn't
have too many rules to write.

> Another conf question - is there any technical problem with having a two
> node direct routing lvs cluster, where the web servers run on the same
> machines as the primary and secondary load balancing routers? that's not a
> configuration i'm planning on using, mind you, i'm just idly curious if
> that's an effective solution for those with low budgets.

That _should_ work fine.

-- 
Horms


<Prev in Thread] Current Thread [Next in Thread>