| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 1011.1 | Cluster alias Address is a bad idea | BIKINI::DITE | John Dite@RTO DTN 865-4065 | Mon May 26 1997 21:43 | 33 |
| > The DSA and MTA run on one CPU only at a time. If this CPU fails,
Why the MTA? The MTA can run on both systems in parallel.
> The transport is RFC1006. We use DECnet alias and Internet alias
> (via UCX).
If you are running over RFC1006 then you can't be using the DECnet alias!
The Internet alias cluster address is a completely different kettle of
fish!
If you know UCX then you will realise that the UCX IP cluster does not
have the same functionality as the DECnet cluster alias.
If you know that the DSA can only run on one node in the cluster
then it does not seem a sensible thing to do to use the cluster alias
address (either DECnet or UCX IP alias address) as you will not be able
to influence where your Connect Request will be passed to.
However, you can add all the cluster members individual addresses to
your dua_defaults.dat so that, assuming one DSA is running on one of your
cluster members, a connect can/will be established ie. X.500 tries the
various addresses in dxd$dua_defaults.dat till if finds one that works.
Now, what you still have to deal with is how to 'restart' a DSA if it fails,
either on the same node or on the alternative node.
Now if you had something like DECsafe for OpenVMS ;-).
Hope this helps.
John
|
| 1011.2 | OpenVMS DSA Failover wish | BIKINI::DITE | John Dite@RTO DTN 865-4065 | Tue May 27 1997 00:11 | 14 |
| This is more aimed at Engineering. I know money is in short supply
and X.500 on VMS may not have such a rosy future as on other platforms.
However one 'small thing' that could/should be done to alleviate the problem
of failover in a cluster is the use of the distributed lock manager.
Why can't the DSA server processes be started on each individual member of the
cluster. The first one reserves an exclusive cluster-wide DSA specific lock
and carries on working in the usual fashion. The other DSA servers then idle
waiting for the release of this lock. In the event of DSA failure, the next
waiting DSA server receives the lock and carries on with his startup etc.
One can but dream....
|
| 1011.3 | decnet alias not used | OSITEL::rtont2.rto.dec.com::Erich | SI-Office | Tue May 27 1997 12:52 | 7 |
| Hi John,
thanks for the input, it was very valuable for me.
DECnet alias: it is present, therefore I mentioned it, but we do not use
it for MTA/DSA transport and agent connect
regards, Erich
|
| 1011.4 | | FORTY2::PULLEN | Julian Pullen | Tue May 27 1997 20:50 | 5 |
| For the record the DSA on VMS does take an exclusive cluster-wide
DSA specific lock. If the lock fails the DSA currentlyt exits.
Julian
|
| 1011.5 | still hopeing... | BIKINI::DITE | John Dite@RTO DTN 865-4065 | Wed May 28 1997 12:45 | 24 |
| Julian,
>For the record the DSA on VMS does take an exclusive cluster-wide
>DSA specific lock. If the lock fails the DSA currentlyt exits.
I realised this, observed by the message below in
dxd$directory:DXD$DSA_STARTUP_OUTPUT.LOG if a second DSA tries to start if
another DSA is running.
%SYSTEM-W-NOTQUEUED, request not queued
Warning: DSA not started, because a DSA is already running.
It's at this point I would have thought one could implement the changes in that
any subsequent DSA server processes would hibernate and idle (ie not be
accessible) and wait for this lock to be released rather than exiting.
In the event of the lock being released the first waiting DSA to reserve the
exclusive lock would then start loading the DIT and establish his channels
to DECnet/OSI etc..
As you may have gathered at present there is no optimal failover solution for
X.500 and OpenVMS.
John (still hopeing)
|
| 1011.6 | | 16.11.160.133::FORTY2::PALKA | Andrew Palka Altavista Directory | Wed May 28 1997 14:12 | 17 |
| The dsa does not currently wait until the previous dsa exits.
However, even if it did that would not completely solve the problem.
You have to issue the ncl 'create dsa' and 'enable dsa' commands
to make the dsa start up.
I believe that it is perfectly possible to write a command procedure
which does what is necessary, using it's own locks. This would start
up the dsa process and issue the ncl commands when it obtains the
lock (you probably need to write a program to obtain the lock. I
dont think you can do that directly from ncl, apart from using 'open'
to poll the status of a lock file).
If you want a solution quickly then that is the way to go, otherwise
you should make your requests known to the product manager for
possible inclusion in some future release.
Andrew
|
| 1011.7 | Product Manager=Nad Nadesan ? | BIKINI::DITE | John Dite@RTO DTN 865-4065 | Wed May 28 1997 20:59 | 10 |
| We will most probably do something by evaluating the DSA DXD$DSA_LOCK.
We will try and make a request to Product Management (Nad Nadesan)
and we'll hope that it's all in Version 3.2 ;-0.
Now that UNIX cluster (Steel?) is coming out with a common file system and
something similar to the DLM what impact will that have on the future design
of X.500 for Digital UNIX?
John
|