high availability? anybody else?

[Rusty Carruth]

Well, here's an interesting question:

> >Hi Oracle Gurus, 
> >
> >I'd like to use your experience in High Availability for Oracle
> >database servers ( specifically on AIX ). 
> >
> >I've been asked by a System Manager to give him some hints on how 
> >too achieve high availability for Oracle. 
> >
> >To put the thing in context, there will be a central database 
> >server for the whole country, around 30 regional machine used to
> >collect data from around 100 000 points. 
> >
> >Data collection will normally be done during the night and I've been 
> >told we can expect 4Tb of data .....
> >
> >Oracle version will be 8.1.6 
> >
> >The central server will be central to the business, it will be used 
> >as well for billing, linked to other big legacy systems with MQSeries, 
> >blah blah blah, ... 
> >
> >So the customer want a very high availability for the system.
> >
> >What I was thinking of is replication. 
> >...
> >
> >The idea would be to have a master Oracle server replicating its
> >data to a read-only slave. 
> >
> >This would bring several benefits:
> >
> >- This would enable the system to have a standby database ready to 
> >take over the master 
> >
> >- The read-only slave could be used as the report server. The master
> >could be optimized for insert and update while the slave would be
> >optimized for read.
> >
> >- The database backup could only occur on the slave to not degrade
> >the master performances
> >
> >- This might as well even lessen the availability figure for each 
> >server. Even if the whole system need a 99.9% availability, with
> >the 2 servers, perhaps the master and the slave would only have to 
> >be 80% available, each servers covering different needs. I think
> >this is very important because whatever the greatest server you use,
> >you might have to upgrade or path the system and this simplify
> >the administration to know that you can have a bigger downtime
> >if needed.
> >
> >
> >Now this is all very nice on paper. 
> >
> >Is Oracle replication really working and not just a hype or
> >too difficult to implement ? ...
> >
> >For example, is it possible for Oracle to break the replication
> >mechanism ( while the master is upgraded ) and then reenable
> >it later ( like this possible with disk mirroring ) ? 
> >
> >
> >The other possible option I see is coming from the experience of
> >a well known bank in France which was already in great financial
> >disaster when their Head Quarter burned, with all their main 
> >computer system with it. It was thought this was the end...
> >
> >Well, not at all, the next day they where able to trade on the
> >stock market as usual. That was because all their data was 
> >mirrored on another center linked by fibre channel. They even
> >got some respect for that for the first time since a long time !!!! 
> >
> >This was some years ago. With the advance in storage since, this
> >might be a good alternative to the Oracle replication with a 
> >price tag that not only banks could afford ...
> >
> >As well any experience for this kind of technology,
> >storage mirrored on another center 10 Kms away but linked by fibre 
> >channel ?  Is it expensive and working reliably ? 
> >
> >Being smart, we could combine the 2 options. 
> >
> >Use the disk mirroring to have a stand by database on another computer
> >center which would be used for the backup as well,  while we use Oracle 
> >replication for the report database to not degrade performance of the
> >slave and lessen the availability figure. 
> >
I'm curious if anyone has done anything liek this with linux?

rusty