TCPIP services (Formerly UCX)

[Swaff]

I am looking for a link to setup TCPIP services. I have a 2 node cluster, each node contains 2 DE60? tcpip cards, the first port on each is enabled. EI0 and EI2. Each connection contains an IP address for node 0, EI0 and EI2 have the same IP. (Is this correct? or do each need a different IP?) I would like to setup a load balance connection to these nodes as well, using the cluster alias TCPIP connection. Is there instructions as to how to set this up somewhere. I have seen many articles on TCPIP, but nothing clear enough. This area seems to be a bit of a void for me.

Thanks

 

[mjpearson]

I don't know and I'm speculating at this point:

It would only seem natural to have two independent IP addresses. If they were the same and connected to two different Switch ports on an Ethernet switch, I would suspect it would cause all sorts of problems. I would even venture to bet that it might even start a "storm".

Unless there were a way to segregate the IP PORT numbers where one IP address would respond to one range of PORT numbers and the other would respond to the rest but this seems like a really odd thing to do.

mike

 

[jlshp]

To properly answer this, we'd need to know the OS version, and the TCP/IP version.

Off-hand, I don't recall if the tcp product can support load-balancing (haven't had a need to set it up, but you could always consult the documentation).

In any case, you can't give both NICs the same tcp/ip address. It just won't work.

To start with, at least setup one NIC on each system with unique tcp/ip addresses.

Then consult the documentation to plan and implement the cluster load-balancing (load broker) features of tcp/ip. Note: This may not do exactly what you want, so read it carefully.

In many cases, you're better off integrating this kind of feature at the router level.

 

[Swaff]

I actually ended up logging a call with HP (digital) and getting an answer. We are running TCPIP V5.1 and VMS 7.3. The answer to my question is
In order to do that you need to use loadbroker which will require:

* a dns server on the network that supports dynamic update,
* A VMS/TCPIP 5.x system running LOADBROKER
* Run the metric server on both cluster members.

Our server will not let us do the update, so, we cannot perform this. I queried my client and was informed that we needed to "failover" and not an Alias. This has saved me much troubles... Here it the procedure to perform that task,
Supposedly TCPIP failover will actually be implemented in V5.4 of TCPIP.

Setup a cluster impersonator as follows:

$ tcpip set int <int_name> /cluster=<shared_ip_address>/c_network=<ip_network>/c_broad=<ip_broadcast>
$ tcpip set config <int_name> /cluster=<shared_ip_address>/c_network=<ip_network>/c_broad=<ip_broadcast>

For example: cluster impersonator address = 10.1.1.3 net mask 255.255.255.0
Cluster members are 10.1.1.1 and 10.1.1.2 same mask interface = WE0

$ tcpip set host mycluster/address=10.1.1.3/alias=&quot;mycluster&quot;
$ tcpip set int WE0 /cluster=mycluster/c_network=255.255.255.0/c_broad=10.1.1.255
$ tcpip set config WE0 /cluster=mycluster/c_network=255.255.255.0/c_broad=10.1.1.255

I want to thank those that responded to my original query.

Harold Swaffield (a.k.a. Swaff )