Re: Advanced Replication: for availability or better performance over distance?

    I ran an advanced replication setup with master sites in the US, Europe and Asia. The purpose was to allow queries to run quicker than they would if say Tokyo users had to run against a server in St. Louis. Multimaster replication would be a bad choice for disaster recovery. You'd have to quiesce replication, generate replication support, etc. etc.

    Conflicts are a potential problem, so you must thoroughly design them out. This might include using a site code as part of the primary key, not allowing users to update data belonging to other sites, triggers to prevent timing conflicts, conflict resolution methods, etc. etc. Start from the assumption there will be conflicts. Design out as many as possible and set up procedures to handle ones that occur.

"Frank van Bortel" <fvanbortel_at_netscape.net> wrote in message news:c41joh$4vl$1_at_news2.tilbu1.nb.home.nl...
> Ernest Siu wrote:
>
> > As subject, what is the main reason for deploying advanced
> > replication? Can someone verify my observation here:
> >
> > Since AR is async,
>
> You decide. synchronous or async are both possible.
>
> if multiple sites are sharing a lot of common
> > tables for read/write, then data conflict will arise frequently.
>
> Which AR can handle.
>
> > Looking into similar products like SharePlex, it seems to me that
> > people use AR or SharePlex more for disaster recovery and high
> > availability than better performance over distances - i.e. few people
> > will deploy active/active servers with AR, instead they use the
> > replica for DR purpose.
>
> That's another assumption. I cannot subscribe to that.
> >
> > If the above observation is true, for nation-wide DB application, the
> > most popular (or the only viable) solution today is one big server (or
> > RAC to some extent) at a single datacenter. Is this how most
> > nation-wide companies (financial, logistics) deploy a geo-seperated
> > DB? DR requirement is another issue that many ways can address
> > (backup, AR, Symmetrix, journaling), but I'm looking more into how
> > people tackle performance issue in a wide-area.
> >
> > Can somebody has some deployment experience on wide-area DB tells me
> > how they address performance and scalability in a nation-wide setting.
> >
> > Ernest
>
> Most common scenario for AR would be just what you describe:
> nation wide coverage, but not all locations need all data.
> You would have central (HQ) data, e.g. the HR system, where
> extracts (exurpts?!?) would be made available to "local"
> branches.
> These branches would enter their (local) data, which, at a given
> time, will be consolidated.
>
> Another scenario would be the traveling salesman: extracts from
> the portfolio are available on his laptop/handheld, new orders
> fed once/twice/x times a day, as well as changes in stock, due
> to other sales.
>
> I have never seen, or heard of, Advanced Replication as a disaster
> recovery insurance.
>
> Unless we're not talking about Oracle Replication, of course.
> --
>
> Regards,
> Frank van Bortel
>
Received on Fri Mar 26 2004 - 18:46:32 CST