|
All of these 'nice to have' features are wonderful, but
the reason the LVS is so great, and so reliable, is that it
is fundamentally SIMPLE. Add a lot of tutti-fruitti gak, and
it will be less reliable, have plenty of failure modes, and
do for LVS what the commercial Linux package companies have
done for Linux. (attempt to make it as messed up as a Microsoft product).
On Fri, 24 Mar 2000, Lars Marowsky-Bree wrote:
> On 2000-03-24T12:21:45,
> Markus Bernhardt <mbernhardt@xxxxxxxxxx> said:
>
> > 1. Load Balancing based upon a hash calculated from src-ip and dest-ip.
>
> Also has the advantage of being "stateless" - you can do hotfailover with
> these without synchronising the state table of two directors (essentially),
> because both directors will make exactly the same decision.
>
> > 2. Load balancing based upon URL
>
> This is being worked on. It is not easy to do though and will not appear
> before the netfilter port to 2.4.
>
> > 3. active/active Configuration of the directors
>
> > Both directors are active and are connected to both uplinks.
>
> Being worked on too. Requires, for the general case, hot failover (so called
> "shared-state") firewall code, which will not appear before Netfilter either.
>
> We are probably going to have an active/standby configuration first and for
> quite some time. "active/active" is a lot more complex. For two uplinks, you
> will also need BGP4 routing.
>
> > remaining one. Perhaps VRRP between the directors would be useful.
>
> VRRP would definetely be very nice to have, though it doesn't give you active
> active either. If someone feels like implementing VRRP, feel free to do it or
> send me a quote for how much you would be willing to ;)
>
> Sincerely,
> Lars Marowsky-Brée <lmb@xxxxxxx>
> Development HA
>
> --
> Perfection is our goal, excellence will be tolerated. -- J. Yahl
>
>
>
|
