Two eth problem!
 

hi all

I have very important and very complicated problem! :)
I have a very small network (5 computer) for cluster.
One of theme is master host and has 2 eth (eth0 and eth1).
eth0 is connected to local cluster lan switch and eth1 is connected to
internet, eth1 should have a valid IP address.

cluster is okey and work good, but when I assign ip address to eth1, both of theme don't work. cluster goes down and i cant use internet.
ping my cluster node and 4.2.2.4 say the eth not work.

what is my problem. and how can I troubleshoot it?
it's really urgent and important!

tanx all

Presumably eth0 and eth1 are in different networks so they need different routes (gateways). You probably need to add a route entry using "route add" command but without more details no one can tell you exactly:

1) What Linux distro is it? (e.g. Ubuntu, Debian, RHEL, CentOS, Fedora, Slackware)
2) What version of the distro is it (e.g. RHEL 5, Suse 11).
3) What is the output of "netstat -rn" and "netstat -r"?
4) What is the output of "ifconfig eth0" and "ifconfig eth1"?

more info about eth
 

distro is debian 5
eth0 configuration is: (lan)
ip 192.168.2.1
mask: 255.255.255.0
gateway: 192.168.2.1
dns is empty or 192.168.2.1 (no need to dns in local lan)

eth1: (internet for remote access and ssh)
ip: a valid ip
mask: 255.255.255.0
dns and gateway come from a dhcp or set manually.

i have no access to computers for about 3 days and i did not try netstat for it. (too bad!)

other computer in cluster lan don't need to have internet access. just master should have.

finally i've tried to assign to ip address to eth0 and (eth0:0 and eth0:1) and connect internet to same switch. but not worked!

What do you do with a switch?
Is there a sixth computer as a workstation and router?
At the moment you can only connect two computer and the internet.

What it should look like:
- 1 workstation with 2 ethernet cards (w-eth0, w-eth1) (or 5 ethernet cards)
- 1 cluster router (gateway) with minimum 5 ports (r-eth0, r-eth1, r-eth2, r-eth3, r-eth4) or four more cards in the workstation
- 1 internet (dsl) router (gateway)

The w-eth0 is connected to the cluster router r-eth0 and w-eht1 to the internet (router, gateway).
r-eth1..r-eth4 is connected each to one cluster computer.
This are two networks one at w-eth0 (or w-eth2..4, cluster-net) and one at w-eth1 (internet)

Keep in mind that a switch can only connect master to one other port and does block the other ports. It is the same like you are running with the cable of w-eth0 in hand to an other machine put the connector in start the network do networking pull out the connector push it in the next machine do networking and so on.

more details
 

sorry, I must describe it more detailed before!

I have 5 computer connected to each other trough a 16 ports switch.
One of theme is master and others are execution hosts. master has 2 NIC (eth0 and eth1) and others have 1 NIC.
master eth0 connected to switch and eth1 is connected to dsl connection.

we need to have ssh connection to master host (to run job on cluster) from internet. eth1 is for this.

O.K. your lan shouldn't have the network 192.168.2.x, choose maybe 2.1.1.x else your master computer is no gateway for the others.
In the moment the internet is reachable from a cluster computer that has been switched to by the master or vice versa. So if that computer has connected to the internet(router) the master and others can not switch any more!
That's why routers or more cards in the workstation (master) should be used.

question
 

i guess you mean i've to change my local lan ip range to another class. is it true?
in fact only master host (which has 2 NIC) should be accessible from internet. (master host controls other execution host and execution host should not have internet access)

Yes, if only the master should connect to the internet then you have only to change your cluster lan to an other network. See /etc/networks
One local network is loopback 128.0.0.0
Normal you can choose any network different from 192.168.0.0 (192.168.routernet.0) maybe 10.1.1.0 , mask is 255.255.255.0 means you can have 255 members (hosts) in that network. It also can be up to 255.0.0.0 .
there are some conventions so you should use the network 10.x.x.x for very local networks, so you get no other problems.

thank you
 

thank you
I'll try as soon as possible and comment the result immediately.