this information is taken from the RAC for Beginners webcast series…..
first some personal notes….
When I first learned what RAC, real application clusters, is I thought I understood it. It is basically clustering of an application on multiple systems. The application is the Oracle database. To some extent this is true. It is a great oversimplification of what it is because it also brings in load balancing, scheduling, resource allocation, affinity, and other attributes typically associated with an operating systems or a distributed processing manager.
- database : set of files that comprise information, this includes metadata, files, and information related to the data
- instance : memory and background processes used to access a database. For RAC you have multiple instances for a database. For Oracle, you can have multiple instances per database but never multiple databases per instance.
- clusterware : component for RAC that takes care of cluster membership including heartbeats, split brain situations, instance managements
- SAN : storage area network. typically collections of disks controlled by a storage area network controller. This is a requirement for RAC
- local and shared storage : shared is disks accessed by multiple hosts. Local implies that only one system accesses the storage
- raw device and cluster file system : methods for accessing data on shared storage
- ASM : automatic storage management, an option for managing raw and cluster file systems available with 10g
History of RAC
- early 90’s – oracle parallel services with v7
- 2000 – enhancements to OPS with 8i
- 2001 – upgrade of OPS with Cache Fusion technology in 9i
- 2004 – oracle clusteraware and RAC update 10g
RAC is not
- set and forget, it does require monitoring and tuning
- transparent to some applications
single instance vs RAC
- single instance: local storage contains instance si1 on storage node A
- RAC: two instances of database (rac1 and rac2) located on two storage nodes. The nodes must exist on shared storage and both servers have a cluster interconnect for heartbeat
- shared database components – control files, temp tablespace, application tablespace, server parameter file (spfile)
- unshared database components – redo logs, undo tablespace, rollback segments. Note that these components reside on shared storage but there are copies of these for each node that are part of the RAC.
licensing issues for Standard Edition
- max 4 CPUS per cluster
- must use ASM for all database storage
- must use only Oracle Clusterware
note that the CPU cound is a limit for Standard Edition, for Enterprise Edition the limit does not apply
- prepare the hardware. it does require multiple network connections and SAN storage. The platforms must be the same and the OS must run the same version. It is recommended that patch levels are the same. With 10g this requirement is a little more flexible and only requires that the platform be the same.
- clusterware – for UNIX, installing SSH key pairs is best, onWindows the username/passwords must be the same
- ASM – this should reside in a separate ORACLE_HOME from the database mainly for patching and downtime requirements. You will need atleast two disk groups, data and flash recovery area. If using for Standard Edition, all database data must be controlled by ASM
- RDBMS – install without database creation. recommendation is to have multiple oracle_home locations across the cluster. OPatch is cluster-aware and will patch all systems in the cluster. Once patches are applied, use DBCA to create the database
- same as single instance tuning, everything still works
- network bottlenecks are the most common issue
- statspack, ADDM, and AWR are cluster aware
- 10g Enterprise Manager has good info
- recovery is more comples because there are multiple sets of redo logs for each node.
- there is just one database, not one per node
- ASM and RMAN are cluster aware, ASMCMD does not currently offer backup commands
alternatives for HA
- microsoft cluster server with oracle failsafe
- veritas cluster server
- polyserve matrix server
- red hat cluster server
- hp mc serviceguard
- ibm hacmp