Two cards on same subnet

[steve_26]

HPUX 11.0. Can someone tell me how the default route is determined and which card will be used for oubound packets when two network cards are defined with the same subnet. The is a situation we will remedy, but I'd like to know for future reference. We made no config or hardware changes but our outbound traffic changed to the other card. I should note the network guys were making switch changes.

[saju_2]

Hi

What is the output of netstat -r -v?

This command shows the routing information in your server.

Regards
CS

[Zeev Schultz]

"Default Gateway for each Physical IPv4 Interface"

http://docs.hp.com/en/5991-0782/ch01s02.html

[Steven E. Protter]

You can not under any version of HP-UX put two cards on the same subnet.

Ex for clarity:

lan0
192.168.0.10 255.255.255.0

host is zero network is 192.168.0

lan1
192.168.0.11 255.255.255.0

host is zero network is 192.168.0

Trying to change to this configuration will collapse network connectivity immediately.

Changing /etc/rc.config.d/netconf to this type of configuration will result in a system that will not boot to the network.

SEP

[Zeev Schultz]

SEP yes you can but its not recommended :) ip_strong_is_model lets the packet to be either sent through the gateway (interface) that received it or with option 2 pick the best one.

[Andy Torres]

Steve, take a look at Auto Port Aggregation (http://docs.hp.com/en/J4240-90031/index.html) and what it offers, if you're looking to have multiple cards in that type of configuration for fat-pipe and/or failover ability. Your network guys could also look into load balancers.

[Wim Rombauts]

As far as I know, putting 2 LAN-cards on the same subnet is unsupported.
But why do you want 2 network cards ?
If you are looking for performance, try the Auto Port Aggregation software.
If you are looking for network redundancy (each LAN-card is connected to another network switch on the same subnet), APA is not usable, MC/ServiceGUard is possibly too expensive, but there are other LAN-failover softwares in the world.
Or you could split your network cards over different subnets. Fofr instance : 1 LAN-card on a subnet that only speaks to other servers. The other LAN-card on a subnet that speaks to clients. This way, you split network traffic over 2 cards.

[Andrew Cowan]

The easiest way to acheive this is logically subdivide your network e.g. if the subnet is 24 bits / 255.255.255.0, then subnet our two cards as /25 (or greater(=). In this way you could address one as .1/25 and the other as .128/25, and as far as your host is concerned they are in two subnets, but all hosts subnetted as /24 would just see two adresses in the same subnet.

[steve_26]

As I stated in my original post, I plan to do away with the conflict. I was just curious about the mechanism. We went to a GigE card some time ago but the 100Mbit card was left active on the same subnet. All in/out traffic was flowing through the gig card until a few days which happened to coincide with the network guys upgrading switches. Right now all inbound is on the gig and outbound on the 100. I was curious why the routing changed. Could this have happened due to some momentary network disruption?

[Andrew Cowan]

The flow of traffic from the host will be controlled by the host, therefore the only way it could be switched externally would be if you were running a routing daemon such as RIP, and your switch broadcast'ed new information.

If I remember correctly HP-UX relies upon the order of the "netconf"? file it starts from lan(0), route(0), then 1,2..

Sorry if this is vague but I don't have a HP-UX box to hand, and so have to rely on my ever failing memory :(

[Steven E. Protter]

Hi Zev,

Mah Shalomchah?

Its been a while since I actually tried it. I did try it on 11i v1 with the March 2002 Core OS and promptly blew up the networking.

I'm glad the effect is not so bad and still don't recommend it.

SEP

[Zeev Schultz]

Steve,like I wrote it may work but most probably will be unstable (ie NFS mounts may go crazy,etc).Thing I mentioned - ip_strong_is_model bounds traffic to specific physical interfaces.I didn't test it in 2 cards-same subnet scenario but according to the theory all your outbound should go back through the interface it came from - in your case GbitE.

SEP,thanks,all is beseder,hope you have fun in NDS :)