Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.server -> Replication w/ read-only snapshots. Ick.

Replication w/ read-only snapshots. Ick.

From: Jim Reynolds <reynolds_at_vu-vlsi.ee.vill.edu>
Date: 1996/12/30
Message-ID: <5a80pe$n63@vu-vlsi.ee.vill.edu>#1/1

We are running Oracle 7.3 Workgroup Server(WGS) on several different platforms: Solaris SPARC, Solaris x86, and Windoze NT. We have been told that the advanced repolication option is not available for the WGS at all, and essentially that we are stuck using read-only snapshots and two-phase commit for data replication purposes.

We feel that we can keep a master site running almost 24/7, and when we can't we're willing to forego any updates(but slaves must still be available for read-only). Even so, I do have a few questions...

Two-phase commit(2PC) is fine if both servers are working. But if one of the machines is down, the transaction is rolled back and not allowed to take place anywhere. This will guarantee data consistency, but is not very fault tolerant. Are there any tricks that can be played with in-doubt transactions that make them update locally, but wait a while if they can't propagate through?

While a snapshot is being refreshed (complete), what happens to queries that try to access that snapshot'ed data? If I have 100MB of old data, will that continue to be available until the new data is completly propagated over? (i.e. no downtime of a table on the slave?)

I know indexes and related objects are not snapshot'ed over, but if I create them on the slave machine, do they stay there between complete refreshes? Take the 100MB table again... I build an index on it on the slave, it gets a complete refresh, what happens to that index?

And, sort of a mix of the two above questions: what happens to the queries on the slave while the index is being built? Are they slooow? Is the old data with the old index used?

I've heard people talk about fast refresh times as quick as 2 seconds. Is that practical if 10-20 rows need to be updated in that time(i.e. if the first refresh doesn't finish before the next, problemsproblemsproblmes)?

I'd be interested in hearing any and all solutions that people use for 24/7 availability and fault tolerant designs on WGS.

Happy New Year!

Jim. Received on Mon Dec 30 1996 - 00:00:00 CST

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US