Re: Dataguard / Archive Logs

On Wed, 2005-06-01 at 21:16, David Sharples wrote:
> wow, you should write a book on this stuff :-)

I've seen to many people writing books from a close distance. It might not be that much fun, I'm afraid. I'm reluctant in starting that. ;-)

> the san would be synchronous writes.
>
> So If i can try to put it simply - it would work that way, but
> dataguards on the face of it seems the better, cheaper solution ?

It depends. If you do not have any Oracle Software running on the standby (at least for 10 days or less per year, that is to allow you to test the setup) you don't have to license the standby. Data Guard requires Enterprise Edition. If you're on CPU-licenses, that costs extra. AFAIK, on Named User plus it's all included. Maybe you can run on fewer CPUs on the standby to save on the licenses. Maybe the standby SAN is more expensive that an extra Oracle license. It will require pretty much the same HW on both ends, making extensions more expensive? What is the network bandwith required and its associated cost? What's the business case in the long run? It really is a trade-off, which I cannot judge from here. The 'value' of delayed applying of redo is unknown, it depends on your organisation, the amount of human errors, their impact on the business.

As you can see more questions than answers. That's what planning High Availability is all about: create a problem for every solution. When all problems have been taken care of (or have simply been accepted as a known risk), then implementation can start.

Personally, I favor Data Guard in stead of SAN replication. But, I earn some money on HA/DG consulting, I might be biased ;-)

Best regards,

Carel-Jan Engel

===
If you think education is expensive, try ignorance. (Derek Bok) ===

>
> On 6/1/05, Carel-Jan Engel <cjpengel.dbalert_at_xs4all.nl> wrote:
> > David,
> >
> > Is the SAN replicating synchronously or asynchronously?
> > Data Guard needs less bandwidth than storage based replication. Most of the
> > times SAN's are configured to replicate asynchrously. Just think of the
> > amount of data that needs to be sent over when storage is replicated: Redo
> > writes (all members), archive copies (maybe also redundant?), the writes to
> > data files, and the updates of the controlfile. Mind that storage based
> > replication is often disk block based, or even disktrack-based! You can
> > imagine what amount of data needs to go through the pipe for that. Compare
> > that with just sending the redo entries with Data Guard. The standby site
> > will take care of applying them, writing to data files and performing the
> > archive job locally.
> >

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