Re: Ambiguities regarding Oracle High Availability features

On Fri, 31 Dec 2004 04:36:10 -0800, qazmlp wrote:

Quick definitions to start the discussion. There is a lot more, literally thousands of pages of documentation, so these definitions baarely scratch the surface.

A LOT of this depends on you understanding the difference between database and instance.

> I keep hearing about the following Oracle features w.r.t High
> Availability.
> - Data guard

Oracle's "failover" technology. Basically for disaster recovery. Very useful to sync to an offsite database that would be used in case the original fails. Mature - been around since Oracle8 in part to consolidate the techniques & scripts that people had created in Oracle Server 7.

Until recently, the failover instance was in standby mode only to sync the database. Logical, rather than physical, standby has made some changes to this.

> - Data Replication

Oracle's "replication" technology. Basically for application distribution. An example is 'corporate database' and 'laptop replica with data subset'. Uses SQL statements to extract the data to be replicated and uses Oracle Networking to transmit the data to the target instances. Can work in multi-master mode and master-slaves mode. Mature - been around since Oracle Server 7 and possibly before.

> - TAF
Oracle's client-side redirection technology. Used in conjunction with Dataguard, Replication, RAC or OPS. Allows the client program access to alternate database instances, potentially without user intervention (hence the term "transparent") if network disconnects at the server side or instance fails. Mature - been around since Oracle8i.

> - OPS
Oracle Parallel Server is the Oracle Server 7, Oracle8 and Oracle8i database clustering technology. Basically for high availability, aside from disaster recovery. Obsolete.

> - RAC
Real Application Cluster is the Oracle9i and Oracle Database 10g database clustering technology. Replaces OPS and adds a lot of scalability capability in terms of more users, more data and more performance. Does to the database and application what OS clustering tries to do to the hardware and operating system.

If you understand SMP computing, one way to think of RAC is: from a database perspective, each server (machine) in a RAC cluster is equivalent to a CPU in an SMP machine that has hot-replaceable CPUs. This is radically different from server clustering.

> - Oracle stream

Oracle's new data extract and distribution technology. Provides a mechanism to extract from log files rather than SQL statements to make things like replication more efficient.

Started with late Oracle8i (coincident with LogMiner?) but did not get moving until Oracle9i. This is very powerful but poorly understood by the Oracle customers I see, and therefore not used near it potential (yet).

>
> I am bit confused about what features are available with Oracle versions
> 9i & 10g.

Look at http://tahiti.oracle.com, Oracle's public documentation, and specifically lok at the "New Features" documentation, which details which feature was introduced in each version.

>
> Also, about
> - interworking/relation between these features For e.g. whether Data
> Replication is used by Data guard

None between Replication and Data Guard. Data Guard uses log-shipping.

> - whether 1 feature is a replacement for an another Oracle feature -
> what features are availabile only for Cluster environment - what
> features can be used for multiple clusters that are located
> many kms apart etc.

RAC replaces OPS. Streams may eventually replace Replication.

All of these are independant of Data Guard which is the 'many KMs apart' failover technology.

Many people think of Clustering as providing Failover - when one machine dies, the other takes over. For database instances, the problems with pure failover include: time to detect, time to fail over; time to start up the processes on the other side of the cluster; need for multi-CPU machines to handle load.

Note that performance and load scalability are not generally part of Operating System or Server Clustering, and are usually handled by getting bigger SMP machines at each node in the cluster.

However, performance and load scalability are part of the base criteria of Oracle database clustering. (Think Beowulf)

Oracle Clustering (OPS & RAC) coordinates multiple instances accessing one database. With Oracle Clustering *ALL* instances can and do access the same data at the same time (the disk is assumed to be adequately protected). If one node (server) fails, the others adjust to pick up the load, but there is minimum impact on people using the application on nodes that have not failed.

>
> Could anybody give consolidated information on all these?
>
> Thanks!

That should get you to the point of being able to read the docco on your own.

lol/FGB Received on Fri Dec 31 2004 - 07:57:10 CST