VMS Cluster alias name with TCP/IP

[Andrew Rycroft1]

Hi,

I have an OpenVMS cluster. We recently configured a cluster alias. It does not seem to work as all connections goto a single node. Is there any way to verify that it is active? I have read the documentation about the load balancing, but this looks like it is dependent on DNS. We do not have a DNS server, only local host name files. Will this prevent the OpenVMS cluster alias from working correctly?

Thanks
Andrew

[labadie_1]

it depends about your tcpip version, it may be named "Failsafe IP". See the very good article on that subject in http://h71000.www7.hp.com/openvms/journal/v2/articles/tcpip.html

[Andy Bustamante]

What version of VMS/TCPIP are you running. TCPIP 5.4 removes support for the traditional cluster alias. FailSAFE IP is the new configuration. You can use failSAFE IP without the starting failSAFE server to emulate cluster alias behavior.

See http://h71000.www7.hp.com/openvms/journal/v2/articles/tcpip.html in the OpenVMS journal for a additional information about TCPIP configuration options.

Andy

[Ian Miller.]

What version of vms and tcpip?

the cluster alias address will stay with a node as long as there are connections.

[Jan van den Ende]

Andrew,


It does not seem to work as all connections goto a single node.


This is an inherent limitation of IP!!

I think you are confusing this behavior with the descryption of DECnet cluster alias behavior.
But, IP being 'generic', it only supports functionality that "all the world" understands, and "all the world" does not understand a (censored) about what clustering is.

And, as most "advanced" non-VMS clustering is just a luxury implementation of NFS, so the node that "has" the disks also "has to" have the connections....

The trick around that to NOT use DECnet, and still achieve some load balancing, is to NOT define a "cluster alias", but some alias name which changes its alias regularly over the various node names (by "metric" & "loadbroker, or by DNS "round-robin".

But the IP "cluster-alias" is just an alias for ONE cluster node (the "impersonator", which can only "fail over" to another node when _NO_ connections are active. (which of course will be the case if the impersonator crashes).

Hope this explains things a bit.

Proost.

Have one on me.

jpe

[Arch_Muthiah]

Hi Andrew,

No idea abt your Cluster env. Assuming you have OVMS cluster in a DECnet Phase V network, and used Phase IV-style address.

Please make sure you have added all the necessary nodes in your cluster as part of the cluster alias. if not, use NET$CONFIGURE.COM proc on "each node" in the OpenVMS Cluster environment that you want to become a member of the alias; not necessary for every member of an OpenVMS Cluster env to be part of the alias and make sure those nodes that assume the alias should have a common authorization file.

If this doesn't help, I am sorry. Wish you get good solutions from our ITRC seniors.

Archunan

[Kris Clippeleyr]

Andrew,

Jpe (Jan vd Ende) has it right. IP cluster alias is just an alias on the IP interface that is grabbed by the first node that boots in the cluster. Only if that node leaves the cluster, another member takes hold of the alias. It's a limitation of IP. The only network stacks that can handle real cluster aliases are DECnet and LAT.
Greetz,
Kris (aka Qkcl)

[Colin Butcher]

If you're trying to achieve a means of balancing incoming connections across the cluster nodes based on server load then you need to be connecting by name (through a DNS) and using the metric server and load broker components somewhere to dynamically update the DNS server to reflect the current preferred target node for the inbound connection.

If you're not using a DNS server this isn't going to work for you, unless you set up a small node to act as a DNS server and load broker (and maybe a quorum node too) - or just add those functions to one (or more) of your cluster nodes - depends how big your cluster is.

failsafe IP is all about having an IP address for a service (not a node) which can move around all interfaces in a cluster (between interfaces on the same node first, then another node if the primary node for that PI address fails). It works best when you have a set of IP addresses, one for each service that your systems offer, and you can then allocate the interface order for each service address by setting up failsafe IP interfaces in the cluster in the order you choose.

Change the way you think about IP addresses - they are no longer per interface, but are per service and they can move around the available interfaces within the cluster.

I generally set up a dedicated (private) IP address for each interface (so I can get to it if I have to) and then set up other IP addresses that users of the various services (eg: Telnet, FTP, application specific and so on) can connect to by address (or by name if they come through a HOSTS lookup or a DNS lookup).

This is a very useful description:

http://h71000.www7.hp.com/openvms/journal/v2/articles/tcpip.pdf

Cheers, Colin.