Setting up heartbeats for MCSG

[Darius Shafie]

I have two HPUX systems HP9000 and HP9001 and each of them have two B5509A NICs installed in hardware paths 0/3/0/0 and 0/6/0/0. The purpose of these NICs is to be dedicated for MCSG heartbeats.

I have configured each NIC in the following manner,

HP9000:
0/3/0/0 10.0.0.100 alias=hp9000-100
0/6/0/0 10.0.0.101 alias=hp9000-101

HP9001:
0/3/0/0 10.0.0.102 alias=hp9001-102
0/6/0/0 10.0.0.103 alias=hp9001-103

One crossover cable is connecting the NICs on paths 0/3/0/0

A second crossover cable is connecting the NICs on paths 0/6/0/0

I have edited the etc/hosts file on both systems and it contains the follwoing information:

xxx.xxx.xxx.xxx xxx.xxxxxxxxx.com yoda
172.20.16.55 hp9000
172.20.16.72 hp9001
10.0.0.100 hp9000 hp9000-100
10.0.0.101 hp9000 hp9000-101
10.0.0.102 hp9001 hp9001-102
10.0.0.103 hp9001 hp9001-103


From hp9000 I can ping hp9001 and 172.20.16.72 and thats ok. But I can not ping hp9001-102, hp9001-103, 10.0.0.102, nor 10.0.0.103.

From hp9001 I can ping hp9000 and 172.20.16.55 and thats ok. But I can not ping hp9000-100, hp9000-101, 10.0.0.100, nor 10.0.0.101.

Is there some special settings to configure NICs to work with crossover cables and communicate to another NIC when connected directly?

[A. Clay Stephenson]

First, I would strongly urge you to put an inexpensive hub in the middle of your heartbeat connections rather than using a crossover cable. The "winky-blinky" lights on a hub/switch serve as very valueable diagnostics and while it may seem that you are adding another point of failure, it's still a good idea because your redundant heartbeat networks protect you from the failure of a single hub/switch.

What your /etc/hosts file contains is largely unimportant. The contents of /etc/netconf are what really matter. Moreover, just looking at your IP addresses it appears that all of your 10.x.x.x IP's are on the same subnet.

[Matti_Kurkela]

Your NIC configuration listing does not look like /etc/rc.config.d/netconf content nor any standard command output I know, so I cannot confirm the settings are really in effect.

Please run "lanscan -v" and copy the output here for us to see. If the B5509A NICs are not visible in the listing, you probably need to add the correct NIC driver to the kernel configuration. (I would need to know the HP-UX version to be able to tell the exact driver name.)

Then run "ifconfig " for each interface visible in lanscan listing.

Which version of HP-UX and MCSG are you using?

I'm not sure you can configure HP-UX NICs to point-to-point mode in a MCSG-supported way. Therefore, the best way would probably be to configure each heartbeat cable as a separate LAN segment with the tightest possible netmask: 255.255.255.252 or 0xfffffffc.

If you don't setup the netmasks explicitly, HP-UX will assume a default netmask, which will be either 255.255.255.0 or even 255.0.0.0 for Class-A addresses like 10.*.*.*.

This will make the routing code assume that it is free to choose either of the crossover cable NICs to reach any of the other host's crossover cable NICs. When you use crossover cables, this simply is not true.

I would recommend settings like this:

First crossover cable:
Network address: 10.0.0.100
Netmask: 255.255.255.252
First host: 10.0.0.101, hp9000-101
Second host: 10.0.0.102, hp9001-102
Broadcast address: 10.0.0.103

Second crossover cable:
Network address: 10.0.0.104
Netmask: 255.255.255.252
First host: 10.0.0.105, hp9000-105
Second host: 10.0.0.106, hp9001-106
Broadcast address: 10.0.0.107

[Steven E. Protter · Accepted solution]

Shalom,

I generally salvage a Cisco switch that networking is one with and avoide the cross connect cable solution.

This will work, either way you go without any special changes.

Note that many versions of HP-UX will not tolerate two NIC card's on the same subnet setting. Networking will suddenly stop if you try this.

SEP

[Darius Shafie]

I agree with everyones recommendations not to use a crossover cable for production servers. However these servers are for a test case.

I found the answer and that is 2 NICs in the system cannot be on the same subnet. I changed the IP address of the NICs on 0/6/0/0 to start with 11.xxx.xxx.xxx and now all works.

Thanks everyone.

[A. Clay Stephenson]

A minor quibble but possibly important: I would not use 11.x.x.x because that is not a private network address and you now might have problems attempting connections to the outside world that do actually use the 11.x.x.x address space. If you want to set up two dedicated heartbeat networks then chose two Class C private network address spaces such as 192.168.1.xxx and 192.168.2.xxx so that you will avoid potential problems if your cluster is connected to the Internet.