| Title: | WinNT-Clusters |
| Notice: | Info directories moved to DECWET::SHARE1$:[NT_CLSTR] |
| Moderator: | DECWET::CAPPELLOF |
| Created: | Thu Oct 19 1995 |
| Last Modified: | Fri Jun 06 1997 |
| Last Successful Update: | Fri Jun 06 1997 |
| Number of topics: | 863 |
| Total number of notes: | 3478 |
I have a prospective customer for V1.1 of Clusters, and unfortunate I'm
SQL illiterate.
They have a small (5GB) Ms-SQL V6.5 database which is critical to their
operation. The database needs to be both available and responsive so we
are proposing that they use MS-SQL replication to replicate the database
onto the second node of the cluster. This database is the 3rd tier of a
3 tier client/server system and is accessed by an application running
on the 2nd tier.
The intention is to use a WINSOCK / TCP/IP connection between the
application and the database and will pseudo-randomly access either the
main database or the replica. Is this a viable approach?
If it is a viable approach, what happens when a cluster transition
occurs and we try to put the main database and the replica on the same
node of the cluster?
Thanks,
Paul (my brain hurts :-))
Tech Support
Sydney
| T.R | Title | User | Personal Name | Date | Lines |
|---|---|---|---|---|---|
| 605.1 | I don't think so | MPOS01::naiad.mpo.dec.com::mpos01::cerling | I'm@witz.end | Fri Feb 07 1997 13:09 | 15 |
I doubt it will work. You will end up with two databases on the backup node with the same name. SQL won't allow that. Even if you could implement it with a different name for the replicated database, the replicated database is read-only. I suppose you could build some fancy logic into your program which 'will pseudo-randomly access either' to know whether this is a read-only request or a write request. My head is beginning to hurt, too, thinking of the mess they are trying to implement. With such a small database, what sort of transaction level are they looking at that would require performance that can't be delivered by a single node? tgc | |||||
| 605.2 | GUIDUK::HEALY | Alan Healy @ZSO | Tue Feb 11 1997 00:27 | 13 | |
There seems to be a lack of experience with SQL replication in
clusters, judging from the notes in this conference. I would be VERY
careful in what you promise the customer unless you know it can be
done. I'm not saying it won't work, but knowing how much stuff is
kept in sync during replication, I would want to see it work first.
By the way, even normal replication is not guaranteed to be 'up-to-the-
minute'. The replication task reads the log of the publishing
database, and I think there is a limit to how often the replication
task can run. If they are trying to maintain an exact copy of the
database they may need to use a 2-phase commit protocol instead.
Al
| |||||