[Info-vax] TCPIP$ logical names

[Phillip Helbig---undress to reply]

As we all know, it seems that cluster support in HP TCPIP Services was
added as an afterthought and/or by someone who didn't know that much
about clusters. 

Probably, most clusters are "shared everything" clusters, where the 
whole idea is that it shouldn't matter which nodes are in the cluster.  
In SYLOGICALS.TEMPLATE the possibility of defining logical names to 
point to system files which are not in their default locations is 
mentioned.  Apparently, something similar is possible for TCPIP by 
defining logical names (and locating the files somewhere else), but this 
is harder to find in the documentation.  I think the shared everything 
cluster should be, if not the default, at least prominent in the 
documentation with respect to all utilities.

Back when I had VAXes in my cluster, I tried this out:

$  IF NODENAME .EQS. "DANEEL"
$  THEN
$    CLUSTER_DEF TCPIP$CONFIGURATION CLUSTER_SYSTEM:TCPIP$CONFIGURATION.DAT_VAX
$    CLUSTER_DEF TCPIP$HOST          CLUSTER_SYSTEM:TCPIP$HOST.DAT_VAX
$    CLUSTER_DEF TCPIP$PRINTCAP      CLUSTER_SYSTEM:TCPIP$PRINTCAP.DAT_VAX
$    CLUSTER_DEF TCPIP$NETWORK       CLUSTER_SYSTEM:TCPIP$NETWORK.DAT_VAX
$    CLUSTER_DEF TCPIP$PROXY         CLUSTER_SYSTEM:TCPIP$PROXY.DAT_VAX
$    CLUSTER_DEF TCPIP$ROUTE         CLUSTER_SYSTEM:TCPIP$ROUTE.DAT_VAX
$    CLUSTER_DEF TCPIP$SERVICE       CLUSTER_SYSTEM:TCPIP$SERVICE.DAT_VAX
$  ENDIF

That is, on one node I defined these logical names to point to files in 
non-default locations (which were just copies of the files in the 
default location for that node).  Not long after that, I replaced the 
VAXes with ALPHAs; until then, the VAXes usually didn't have the TCPIP 
cluster alias, so they didn't actually do much with TCPIP (maybe just 
the few tests I did).  It seemed to work OK, but of course just one node 
was actually using these "common" files.

Are these all the relevant logical names?

Is anyone actually using this feature?

The idea is that since as much as possible is the same on all nodes, it 
would be nice a) to have it all in one place and b) hence necessary to 
change it only once, from any node, rather than keeping all nodes in 
sync per hand.  Do the logical names above refer only to such 
information?  (Presumably, one could also move node-specific stuff, but 
I don't see much point in that.)

In September or October, I plan to upgrade to VMS 8.3 and whatever TCPIP 
is on the distribution disk.  Since this upgrade (from 7.3-2) is 
apparently not as painless as other upgrades have been, I plan to read 
all the relevant documentation and go through things step by step.  I 
think it makes more sense to reconfigure stuff on the cluster before the 
upgrade and make sure that it works since otherwise I have to 
reconfigure it long after the upgrade, which would mean continuing to 
configure all nodes separately (including new features), or reconfigure 
it at approximately the same time as the upgrade, which makes 
troubleshooting more difficult if something doesn't work as expected.