LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Re: More on scheduling...

To: Benjamin Lee <Benjamin.Lee@xxxxxxxxxxxxxxxxxxxxxx>
Subject: Re: More on scheduling...
Cc: lvs-users@xxxxxxxxxxxxxxxxxxxxxx
From: Julian Anastasov <ja@xxxxxx>
Date: Wed, 25 Oct 2000 09:00:03 +0000 (GMT)
        Hello,

On Wed, 25 Oct 2000, Benjamin Lee wrote:

> On Wed, Oct 25, 2000 at 12:43:55AM +0000, Julian Anastasov wrote:
>
> >     Yes but LVS uses different connection table optimized for
> > the client addresses.
>
> Ah, I see.
>
> >     The LVS forwarding methods are same in 2.2 and 2.4. LVS uses
> > only the Netfilter framework, i.e. the hooks but nothing from the
> > connection tracking. Currently, you can't run LVS with ip_conntrack.o
> > but this is still possible with ipchains.o!
>
> Would there be any advantages to using the ip_conntrack.o vs. the current 
> "connection table optimized for clients"?

        There are some hacks in LVS that are not possible to implement
in the netfilter environment. The other thing is that currently the netfilter
framework does not handle the case where two or more conntrack modules can
coexist: ip_conntrack/nat work at pre_routing while LVS runs at local_in for
many reasons: fwmark support, direct routing, etc. This is the main problem
in the 2.4 implementation: LVS can't coexist with ip_conntrack/nat. But
it is stable enough with ipchains.o

> Cheers,
> Ben.


Regards

--
Julian Anastasov <ja@xxxxxx>



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