Re: active_instance_count behavior

Hi,

AFAIK, the only two components which know about INSTANCE_ROLE are the instance itself and the listener. The later is informed about this fact by corresponding PMON process.

In any other respect the instance with INSTANCE_ROLE=STANDBY_INSTANCE is no worse than that with PRIMARY_INSTANCE role. You can run jobs on the standby instance, you may fill library/row cache with the objects from primary instance (DBMS_LIBCACHE). You event can allow ordinary users to connect and work with standby instance.

The listener will show the status of the service as BLOCKED, thus preventing connections to it. However you have 3 ways to walkaround: 1. Local connections (no listener involved) will be successfull 2. Use static configuration in listener.ora, use (CONNECT_DATA=(SID=...)) in tnsnames.ora (i.e. <=8.0 syntax) 3. Use (CONNECT_DATA=(SERVICE=..)(INSTANCE_ROLE=ANY)) in tnsnames.ora

Thus why any static remastering should occur (dynamic one doesn't work in 9i anyway). On the other hand, in 10gR2 there are chances that dynamic remastering will change the mastering responsibility in more reasonable way.

BTW, are you really ready to pay 50% more for licenses with a kind of 50% less processors in use just in order to save few minutes max (say, for creating SGA) on instance restart? I mean, you not planning to use second instance at all? I think, that designers of this "solution" just made it quick & dirty with a thought similar to "why not do it, if it is so easy?" and "marketroids will be happy". ;-)

HTH
-- Andrey

John D Parker wrote:
> Ok, in a 2 node RAC 9.2.0.5(yes, I know, don't ask). We have an
> active/passive configuration with active_instance_count=1 set in both
> instances. This makes instance prd1 the primary. My question hinges
> around behavior of block mastering and cache coherency. In normal
> active/active RAC, each node "masters" approx half the datablocks.
> Does this behavior change when I have a primary node? I would think i
> would. However, I can't find any docs one way or the other. With a
> Primary node and blocks being mastered like in active/active, the
> primary node would be continually asking the passive node for blocks
> which seems very inefficient. It would make more sense for the primary
> node to master all the blocks and behave much like a pre 8.1.7 OPS
> node. Any thoughts here? I'd really like to see some docs one way or
> the other but I can't seem to locate any.

--
http://www.freelists.org/webpage/oracle-l