Alexei_Roudnev wrote:

> Common rule is:
> - RAC is protection against short failures (server reboot, you need to
> insert new card, RAM error,
> you need to patch Oracle, etc);

Yep. Provided your client has the money to invest in RAC plus DataGuarg
(3 servers, one shared storage at least, RAC licences...). From
experience, even handing out the bill for a second backup server leads
to endless discussions about the necessity to have the exact same
configuration as production for a standby server.
On the other hand, if the 10gR2 fast switchover feature works, DataGuard
could comply to some extent to this kind of event, too. Plus you don't
need RAC licences, shared storage, etc... We would still have a short
downtime, though. Anyone implemented that feature yet? Any catches or
pitfalls?

Actually, it's not quite clear to me how many Oracle licences you should
have in a three-node DataGuard configuration, where the first standby
server on your LAN, and the second on a remote site, and both standby
servers are only there for failover or temporary switchover due to
maintenance on the main server.
In a two-node configuration, it is "clear" (if you read the small print)
that you need only one licence if you don't use the second server for
anything else than failover. But when you've got two standby servers...?

> - DataGuard is protection against long failures (disk array failure, power
> failure in the building - if you have remote location,
> DBA made error and removed important tablespace, etc etc).

Or leaking old water pipes upstairs, people drinking coffee in the
server room in midsummer because it's the only air-conditioned room in
the building... ;-).

> So, with RAC, you wil not see short (5 - 30 minutes) downtimes which you
> usually have to maintain system, or when system panic, reboots and so on.

Except if RAC panics due to a software bug.

> With DataGuard you wil not see long (6 hours) failures which you can see if
> database is damaged, disk ssytem die, DBA make mistake during upgrade and so
> on (I did not see it for a long, but such events happen time to time).
> No one system with shared _something_ (disks for example) and with 100%
> syncronized state (RAC) can protect you for a long-turm failures. Any disk
> system can fail, or important data can be damaged easily, or software bug
> can kill RAC cluster (very likely event, btw).

That was my point. Too often management gets a woody on market speak
over automatic switchover, versus real data protection. On the other
hand, here in France, criticizing management decisions can be a valid
firing cause ;-).



--
To unsubscribe, email: suse-oracle-unsubscribe@(protected)
For additional commands, email: suse-oracle-help@(protected)
Please see http://www.suse.com/oracle/ before posting