|
At 11:48 PM 6/13/00 -0700, Horms wrote:
>On Wed, Jun 14, 2000 at 02:48:12PM +0900, ??? wrote:
> > I see.
> > Then the only problem is that LVS-DR only receives (and routes) packets
> > whose destination is itself.
> > If LVS-DR takes every packet and selects an ethernet device to be used
> > for routing, then it may be used to duplicate(to give high-availability
> > or to balance load) network devices such as switches, routers, firewalls,
> > etc.(It is the functionality of L4-switches as I know)
>
>My understanding is that the term layer 4 switching refers to the direction
>of traffic based on layer 4 of the OSI 7 layer stack instead of layer 2.
>This is what LVS does.
>
>The additional functionallity that you speak of is more to do with the fact
>that a switch - being part of the switching fabric - sees all packets for
>hosts that are plugged into it. A host cannot do this. We can, however,
>achive high availability by monitoring other hosts and using methods such
>as ip address takeover and utilising interior routing protocols such as
>OSPF.
Another thing that switch does LVS does not is that switch can send
packet between the ports based on MAC address, rather than routing
the IP packet based IP header. That makes layer-4 switch constantly
outperform router based solution in the lab testing. However, no one
had done any real life test between layer-4 switch and router based
load balancing solutions to compare the performances. The point of this
is that in the lab, layer-4 switch can see all the MAc addresses of all
the simulated clients and servers, so it can route the packets based
on that. LVS can not do that -- no special hardware for that.
|
