Re: Very Large OLTP database: to or not to?

Nabil Courdy wrote:
>
> The model works perfectly for us since if one region is out due to
> some reason, all other regions continue to work fine. We have
> implemented high availability solution in all regions, this includes
> node fail-over, dual everything (network cards, power supplies, etc.),
> disk mirroring, etc.

This probably cost a fair amount to setup, yes?

> We feel that although Oracle offers nice features such as RAC9i,
> Standby databases, partitioning, etc. The concept of single database
> is still a single point of failure at the database level. Surely
> there may be situations when the primary and standby databases are
> out.

Yes, but nothing (but $$$) is stopping you from adding a second standby database. So what are the odds both standbys *and* the primary are out? OK, so add a third standby... keep going until you're comfortable or broke or hit the standby limit (six I think?).

> Currently, we generate giga bytes of redo logs (journal files in
> Ingres), these files would have to be constantly shipped to keep the
> standby database sychronized.

That's one way of transferring the information (i.e. gigabytes of archive logs at a time). But Dataguard can also be configured to ship the redo info via the log writer process in smaller, more frequent increments than single archive logs. There are pros and cons of course.

But basically is comes down to what the business can tolerate. If the business can live with one or two of the regions going down at a time (not happy about it, but could live with it), but never tolerate all 6 regions down, then you probably don't want to go the single database route (in that case, maybe consider Oracle Replication?). If the business can tolerate less frequent and shorter outages of all six regions, then a single, highly-available database might make sense. Must factor in cost of maintenance for 6 vs. 1, hardware, support, etc. etc. And don't forget upgrades/patches (hardware, OS, Oracle) - you'd have to think seriously about how you'd tolerate outages for this sort of thing in both scenarios.

No easy answers.

Regards,
Sean Received on Thu Mar 28 2002 - 11:30:07 CST