Redundant network connection

[Michael Kloth]

I'm running a Alphaserver-cluster/SAN with VMS V7.3-2. All servers have currently 2 network-adapters to have a redundancy for the cluster-connection.
What I'm looking for is to have a redundancy in the network-connection, an additional network adapter which is connected to a different switch. If one switch is done (happened in our location a few days ago), to have still a network-connection at least for TCP/IP (we are also using DECnet IV, but a broken connectivity is less critical).
What I found in the documentation are the functions LAN failover and failsafe IP. But what I understand from the documentation, is that these 2 mechanism help you when a network adapter breaks.
Is it possible that VMS also recovers that a network link to a switch is broken and uses directly the other network adapater (of course with the same IP-address) ?

[Willem Grooters]

What I unserstood of failsafe IP:

Each NIC will be 'touched' over the network once in a while and ought to return a reply. If a reply is not received, the NIC is set to be oofline and faiul-over will take place.

That would mean that is a switch breaks, the 'failing' NIC is taken offline and it;s IP-address will pass to another - if set up that way, eventually to another node in the cluster.

IIRC the same is true for LAN failover, though not clusterwise.

One thing to keep in mind: If you ahve two NICs on a system, there might be a problem with Failsafe IP: you may loose ALL cennections.

[Volker Halle · Accepted solution]

Michael,

LAN failover protects the system from a failure of the LINK between the network adapter and the port on the switch. If the LAN adapter gets a LINK DOWN, it will automatically use the other LAN interface in the failover set. Failover is very fast and will work for ALL network protocols.

TCPIP failSAFE IP only protects your IP protocol stack. failSAFE IP monitors the received byte counter of the LAN interface and declares an interface 'failed' if that counter does not increment anymore. Your IP address can fail over to another interface on the same node (does not disrupt IP sessions) or even to another interface on another node in the cluster (sessions need to be re-established). Failover is not as fast as LAN failover. This is well described in the HP TCP/IP Services for OpenVMS Management Manual Chapter 5

http://h71000.www7.hp.com/doc/83final/6526/6526pro_contents.html#toc_chapter_5

Volker.

[Robert Gezelter]

Michael,

"Failed Adapter" should generally be read as "a loss of connectivity through the path as seen by that adapter".

DECnet IV will route around failed circuits, but it must have the routing capability enabled (which may, depending upon your licensing) involve additional costs.

- Bob Gezelter, http://www.rlgsc.com

[Heinz W Genhart]

Hi Michael

a further advantage of FailSafe IP ist, that you have a higher bandwith. In case you configure 4 NIC's on one machine the network traffic will use all those 4 interfaces, what gives us a higher bandwith.

If you are using LAN failover, only one Interface will be active at any time.

We are using FailSafe. We have many two node clusters. Each node has two DE602 NIC's. Each Machine has 4 addresses. If a NIC fails, those adresses will be transferred to another NIC on same machine. Those 4 adresses are defined as aliases in DNS and the DNS Server is using a round robin algorythm to supply this adresses.

We then have further 8 Adresses for the cluster alias (two node cluster). Those adresses are also in DNS and the DNS server supply those adresses using a round robin algorythm too. If one NIC with a cluster alias address fails, then this adress will be transferred to one of the another 7 NIC's.

And for sure, each NIC port of a Machine is connected to 4 different switches.

If required I can attach a Excel for configuring FailSafe. We used and are stil using tis Excel for every new FailSafe setup.

Regards

Heinz

[John Abbott_2]

Michael,

Just to add some more info.

LAN failover is not supported on some of the older alpha network cards, so check the documentation.

failSAFE IP; I remember from a TUD that on quiet networks a phantom failover may happen on systems set-up with only two NICs operating on failsafe mode, so three are recommended.

J.

[Andy Bustamante]

As fare as LAN failover, one of LAN ECOs, now included in the latest UPDATE ECO added support for the DE-500 NICs, otherwise see the System Manager's Manual.

I've configured failSAFE IP on LAN failover (LLA) connections. We use LAN failover since we still have DECnet phase IV internal applications. End users access a "TCPIP service address" which can migrate from node to node in a cluster.

We upgraded switches at one site one night with no downtime, connectivity just failed over. By using at least two switches, you can have one fail and maintain service.

Andy

[Michael Kloth]

Thanks for all of your replies. Now I understand more clearly how this "LAN failover" and "failsafe IP" works and how I can use it.