HACMP 5.1 standby network addresses

[ggitlin]

Hello,
I am setting up a new HA cluster. Our network engineers are telling me that we only have one IP subnet, so I can not assign the standby addresses from a different subnet, as per HA manuals. Would this present a problem for HACMP configuration - in other words, does HACMP allow this?

Thanks in advance.

 

[dlpar]

It's a real problem.
2 solutions:
- Ask your network engineer for a second VLAN.
- Use two different mask(subnet) to split your VLAN

 

[ggitlin]

Thanks, dmarais,

When you say "It's a real problem" did you mean that the cluster will not work or it will, but it is an incorrect way of setting it up?
In regards to the second suggestion, I found the following in HA redbook (an old one):
"
The netmask for all adapters in an HACMP network must be the same even though the service and standby adapters are on different logical subnets.
"

Thanks in advance...

Greg.

 

[sbix]

From a theoretically view point there is no need for a different subnet.
You could use also addresses of the same subnet for standby adapters, if you have some free addresses ....
It's not reccomended ... but .. you can try

 

[bi]

But if you use the same subnet you have introduced a single point of failure, right?

 

[sbix]

It depends ... the most of the times adapters are connected to the same hub/switch, so it's a well known SPOF usually tollerated by network engineers or customers

 

[PSD]

The key point here by having the service and standby interfaces on the same subnet is that HACMP cannot explicitly define which ethernet adapter is being used during heartbeats. This will cause problems when working out network adapter and network failure. If you have two different subnets then each one can be used explicitly for heartbeating and will enable HACMP to isolate the failure.

Use two different subnets.....and two differnet network switches to remove SPOF from the equation.



PSD
IBM Certified Specialist - AIX V4.3 Systems Support
IBM Certified Specialist - AIX V4 HACMP

 

[sbix]

Are you sure?
Once ago I tracked the clstrmgr communications using cldiag.
It seemed there wasn't this problem ... but it was an impression

 

[PSD]

sbix,

Well in AIX 5.1 for example it uses something called multi-path routing so if you have two adapter on the same subnet in the routing table you will see that they are linked. It then starts to round robin between the two interfaces....this is not ideal in an HACMP cluster.

The restriction is not HACMP but routing and subnets, if you have two addresses on the same subnet you cannot guarantee which adapter it will use, the word here is gaurentree, I am sure that in many cases during testing you see the correct behaviour.

In AIX 5.1 though, this is a big no no.



PSD
IBM Certified Specialist - AIX V4.3 Systems Support
IBM Certified Specialist - AIX V4 HACMP

 

[ggitlin]

Appreciate everyone's responces. The OS is 5.1, so it looks like I need a second VLAN. Thanks again for valuable info.

Greg.
AIX Administrator.