The Question is:
We are running a 2-member Cluster as a advanced server 7.3. Our PC-Clients are
mostly w2k, but some still Win95. All PC's connect now with
tcp/ip only. We have Wins-Servers running on two w2k-Servers. Most shares are
connected via the pathworks alias. The cluster has a vmscluster alias also. It
seems that the failover of pc-connections from one other to the other in case
of chrash or shutd
own does not work any longer. (it was working fine when we used an older
pathworks-version and the additional NETBUI-Protocol).
I found some articles on this problem, but none of them solved my problem.
What do i have to set to make it running ?
The Answer is :
There are two general types of cluster failover when considering
client connections made to the cluster alias.
The first type is for broadcast-initiated connections. For these
connections, the Advanced Server package has internal algorithms and
a round-robin scheduling scheme to assure that the load is balanced
across the cluster. As for the expected failover processing with
this environment, the client will re-broadcast the connection request
and the remaining nodes will receive the request and (one node) will
It is likely that when you observed failover to be working, it was
working because of this broadcast scheme. The use of NetBeui as a
transport typically confirms this configuration to the case.
The second type of failover is for directed-datagram-initiated
connections. In this situation, it is necessary to have something
external to the Advanced Server resolve the cluster alias NetBIOS
name to a TCP/IP address. In your case, external resoultion likely
involves WINS, though a client-local lmhosts file is another
possibility. To achieve failover, you need to add static,
multi-homed entries for the cluster alias to your WINS database.
With these entries, connection failures will trigger a connection
to the next node within the list of nodes defined for the alias.
One of problems with this WINS-based scheme is the lack of dynamic
load balancing. While some sites have obtained a level of static
balancing using multiple WINS servers -- each one with a different
ordering of the addresses associated with the cluster alias -- the
OpenVMS Wizard generally prefers to avoid this WINS-based approach.
For true dynamic failover and for dynamic load balancing, the OpenVMS
Wizard typically recommends the use of the IP distributed name
service (DNS/BIND), combined with the use of the OpenVMS TCP/IP
Services product's Metric Server and Load Broker. A detailed
description of implementing this scheme is included within the
Advanced Server for OpenVMS Administrators guide.