Alexey
There are a couple of options here that have been mentioned by the
others. I thought I'd just round them up and add a couple more.
Failover configurations with Oracle take five types, not all of which
require clustering. These are:
- Oracle Fail Safe - runs only on NT in an NT cluster configuration.
Here one node is active and the other is not. If the first node dies,
the second one automatically takes over the database for you. Very
simple, not very expandable since NT doesn't support a lot of nodes
natively.
- Oracle Parallel Server - in this configuration, you have a cluster
using shared disks. The database is located on the shared disks, and
there are separate instances (background processes + SGA) on each node
of the cluster that is used by OPS. More complex to manage (although
this is becoming easier with the 8i releases with the use of cache
fusion), higher performance possible (depending on application
suitability), higher availability since the database can be accessed
from multiple nodes at once. This can also be configured in an
active-passive config if you desire. You can automate the failover at
the user level from 8.0.5 (I think that was the initial release) in a
variety of ways (no failover, connections pre-established, rerun selects
automatically). See the OPS doco (excellent manual) for more details.
- A related config that is not dependent on Oracle clustering
capability is a cluster where the OS handles the failure. This usually
requires scripting the restart of the database on the second node of the
cluster, so it is lower availability than OPS (where the database is
already accessible from both nodes).
- One that may not meet your needs, but still a way of failing from
one site to another, is standby database. In this, a standby database
is created on another machine, and the redo logs are shipped
(automatically in 8i) to that machine and applied to the standby
database which is kept mounted in recovery mode. In 8i the standby can
be started in read only mode if needed then taken back to the mounted
state.
- Again an option that may not meet your needs is replication. In
this case, rows are replicated from the primary database to another
database on another machine. Both databases can be updated at the same
time, so you need to look at conflict resolution if taking this
approach.
Finally, you may be able to achieve what you want simply by having your
tnsnames.ora file point to a primary then a secondary address.
HTH.
Pete
Alexey Raschepkin wrote:
> Does Oracle server support failover configurations ?
> I mean some kind of software cluster or the like ?
Received on Mon Dec 20 1999 - 11:35:17 CST
