On Tue, 2007-09-25 at 13:17 +0200, Jan-Frode Myklebust wrote:
> I see the direct routing docs describe direct routing as a asyncrouns
> connection where packets are coming in trough the LVS and responses from
> the real-servers are routed independently of LVS.
Correct. It would be more accurate to call it "Direct Response", as the
realservers respond directly to the clients without the assistance of
> So what's the problem with having the LVS be the default gw of the
> real-servers? Is it just to avoid being a bandwidth bottleneck, or are
> there other issues ?
Bandwidth is one problem, but nowadays a fairly small one.
The more common problem is that load balancing is often a "retrofit"
onto existing systems, and DR allows you to achieve your aim with little
or no reconfiguration of (what become) the realservers.
It is analogous to many commercial options which place an "appliance"
into a network to do the directing job - Cisco, F5 et al.