Oracle FAQ | Your Portal to the Oracle Knowledge Grid |
Home -> Community -> Usenet -> c.d.o.server -> HA & Failover options
Hi all,
Let me give a brief background before i pose my questions:
Background:
We are currently working on the architectural phase of a program which
will involve implementing a given solution for multiple companies. We
have proposed to have physically separate organisation specific
instances for all the technology components (App Server, DB Server,
Security Infrastructure, Web Server, Infrastructure services for
Systems Management, Backup & Recovery). Major softwares used are Oracle
Application Server 10g Release 3 and Oracle Database Release 2. The
project has some strict HA & Failover requirements
The OS choices are AIX 5.2/5.3 and RedHat Linux 3/4. The box choices are a) Commodity Servers: IBM x Series or HP ML/DI Series with Windows 2003 Server or Red Hat Enterprise Linux ES 3.0 or 4.0 b) Unix Server IBM P570 Series with AIX 5.2 or AIX 5.3.
The HA software choices are IBM's HACMP/XD, Veritas Global Cluster Software, RAC and Data Guard (DG), Oracle Application Server's Clustering features.
Constraints:
Questions:
a) For database which is the best cluster solution: HACMP/XD or VCGS or
RAC+Oracle Clusterware+Dataguard? I'm aware that in hardware based
clustering, the passive instance doesn't get used whereas in an Oracle
RAC solution it does get used. But i don't want to make use of
mid-range servers (p570 is one) along with Oracle RAC and a hardware
cluster since my cost shoots up. I would preferably use commodity
server clusters when deploying an Oracle RAC solution. Share your
thoughts on this. Also, how stable is Oracle clusterware? Please give
your estimation of the three options.
b)Do you have any large scale implementation experience of Oracle DB on low end boxes such as IBM x Series/ HP ML/DI Series with either Windows 2003 Server or Red Hat Enterprise Linux ES 3.0 or 4.0? If yes, share your experiences.
c) n+1 vs n+n
Currently we are proposing a 2*1 instance configuration across two locations for the database based on it being a middle-path solution. i.e., it is not exhorbitantly priced and at the same time it offers reasonable failover and HA facilities. We are aware of the problems in this option such as no protection from instance or node failures at the secondary site, potential performance degradation at the secondary site, manageability problems owing to asymmetric configurations across primary and secondary sites
Can you provide your thoughts on n+n or n+1 configurations when dealing with multiple database instances?
d) Generic vs Specific clustering solutions
Veritas Global Cluster Software has the advantage of being platform independent whereas HACMP doesn't have. Does this factor ever play a role in deciding the cluster software?
d) Cost
>From a cost perspective out of the three options (IBM's HACMP/XD,
Veritas Global Cluster Software, RAC and Data Guard) which one is the
cheapest and which the costliest?
e) When using Data Guard is there a requirement for Global clustering software or will the local clustering software itself perform the job? Also, how instantaneous is failover to the secondary site with data guard when failures occur in the primary site?
e) Data replication
For data replication across primary and secondary sites we have options such as a) SAN or NAS based disk replication b) Data Guard. What are the advantages/disadvantages in these two options?
f) Oracle Application Server
Historically we haven't used Application Server's clustering features in this shop. For load balancing we have used Network Load balancers. We didn't use Application Server's clustering features in the past because the earlier releases of Oracle Application Server had some bugs in this aspect. What is current status? Received on Tue May 09 2006 - 06:35:37 CDT