Guys

I have finally implemented a fully functional RHEL cluster. I used HP ILO device as fencing device and all things are working fine.

I need to know , however, what are the ways of having additional heartbeat mechanisms in RHEL cluster. Right now , i have my ethernet card as SPOF and it's failure are automatically promoted to node failure.

I dont want to use ethernet bonding as my network admin is not confident about ethernet chaneel bonding support on his network swtich.

Is it possible to configure additional ethernet card ( present on my cluster nodes ) as standby interface so that when service ethernet goes down , service ip address automatically moves to standby interface ( not through ethernet bonding )?

Please advice!!

Polani

linux bonding supports many modes, the only mode that your network admins need to configure for is mode 4, which is IEEE 802.3ad / EtherChannel load balanced bonding. other modes are switch agnostic, and are what you're after. you can spit ip packets out of either nic without the network complaining based on many different algorithms, or you can literally enforce a master / slave mechanism, but this won't give you an advantage over a basic LB mode. just check the bonding howto for some more background info.

I have also tried installing a two node Linux cluster (for computational requirements , but not a high availability cluster).

Environment:

1) OS RHEL 4 release 3
2) Networking Typically TCP/IP for communication with two NIC cards
3) Cluster software MPICH2

After completely deploying the mpich2 application I have run a simple c program to chek whether really the program is distributed across the cluster for computation. But I didnot experience this. I found that the time taken is exactly double amount of time as it takes on a standalone machine.

My assumptions:

1) If I run execute program on the cluster, the mpich2 application distributes the computation across the all nodes of the cluster. If so why did my sample program take more time to run?

Any suggestions.

Thank you
Mahendra Nag Jayanthi.

please don't hijack threads. this clearly has nothing to do with the OP's question other than the word "cluster"...

I am sorry for this ..but I could not find a link to post a new thread..any ways do you have any suggestions for my post?

You can use Private interconnect on second interface rather then using router for the same purpose. Private interconnect will work nicely for you as you would need the CAT5 cable to connect the two servers on the secondary NIC card. Keep the IP addresses like 10.0.0.1 and 10.0.0.2.

Oracle provides a hangcheck module which allows the server to listen to heartbeats and respond in case of one node failure.

Rahul Khare