Hi All,
I don't really have too much TCP/IP background (comming from different
proffesion), so my claims may be off. I have setup simple load
balancing for local network to squeeze some more juice out of our old
100M switch. Situation is following Fedora Core 2 server with 1G and 3x
100M NIC and local windoze clients. My ip ro ls looks like this
192.168.3.0/24 via 192.168.3.1 dev eth0
192.168.1.0/24
nexthop via 192.168.1.1 dev eth1 weight 1
nexthop via 192.168.1.2 dev eth2 weight 1
nexthop via 192.168.1.3 dev eth3 weight 1
127.0.0.0/8 dev lo scope link
default via 123.123.123.123 dev ppp0
Where subnet 3.0 is giga and subnet 1.0 is 100M. Up until upgrading to
2.6.8.1 kernel it was working ok, but now I've noticed that if I
want to add default route on the 192.168.1.0/24 class it refueses me all
the time:
[roor@fc2]#ip ro add default via 192.168.1.100 dev eth1
RTNETLINK answers: Network is unreachable
and it won't work in any combination until i delete all "nexthop" routes.
What could that be?
I've read alot around web, regarding load balancing and lots of experts
claim that you can not be better off using multiple NIC on same subnet,
but I think that you can be and our network is definetely faster. One
client is not much better off, but 10 clients asking data from server
are definetely better off, at least it appers so by running a few quick
tests.
I would really like to hear from some guru, if there is some ground in
this. I've also noticed that Solaris has a trick called interface group
which is suppose to do just that - load balancing (ifconfig .. group).
I'm not sure if such thing exists in Linux, but assume that iproute2 and
and "nexthop" routine was done for a good reason.
I think that samba, also lives hapilly with multiple interfaces on same
subnet, but some windows programs experience twiches like "network name
not found", "network file not found".., but mostly seem to work OK.
Any clues?
Thanks
Noerr De Dykul
|
Similar Threads
Linear Mode
