The key distinction between clusters and grids lies mainly in the
way the resources are managed. In the case of clusters, a
centralized resource manager performs the resource allocation, and
all nodes cooperatively work together as a single unified resource
running on the same operating system and hardware architecture. The
result of such aggregation is to present a single system image
(SSI). Clusters
are generally built for a specific purpose; for instance, to host a
parallel database server or for hosting an application server. In
the case of grids, each node has its own resource manager and does
not aim for providing a single system view. They, in turn, provide a
pool of resources for a variety of users and applications.
Grids can span across single or multiple organizations and data
centers. Grids encompass a bigger framework and provide wider and
loosely coupled aggregation of servers and other related resources.
Grid architecture employs specialized scheduling software GME (Grid
Management Entity)
that identifies available resources and allocates tasks for processing
accordingly. Requests for resources are processed wherever it is most
efficient or wherever a specific function resides. Computers or nodes
located in the grid are able to act independently without centralized
control, handling requests as they are made and scheduling others. If
one set of resources is not available, they will simply use another.
Introduction to Cluster Technology
A cluster consists of servers linked via a network interconnect that
present a single system image
(SSI) to users and applications.
Clusters
are used to provide HA (high availability) and load balancing. RAC
(Real Application Cluster) is a clustering solution for Oracle
databases. The software
glue that holds the Oracle 11g RAC together is called Oracle
Clusterware.
Clusterware coordinates database operations between servers so
that the SSI is maintained.
Clusterware is required on all Oracle 11g RAC implementations,
regardless of the operating system or additional cluster software
being used. Any
additional third party cluster software must be certified for RAC.
Oracle Clusterware
Oracle clusterware manages node membership, group services, global
resource management, and high availability functions.
Clusterware is a prerequisite to the Oracle RAC installation.
A RAC database is managed by clusterware.
Other Oracle processes that are managed by clusterware are the
VIP, Global Services Daemon
(GSD), the Oracle Notification Services
(ONS), and the listeners.
These processes are started automatically by clusterware and restarted
when failures occur.
Clusterware is installed from one node (server), but the clusterware
daemons
are installed and run on all cluster nodes.
Clusterware essentially manages all cluster-aware processes.
These processes are called CRS resources in 11g.
The resource settings are stored in the Oracle Cluster Registry
(OCR). The main clusterware resources are listed here.
-
Database
-
Instances
-
Services
-
Listeners
-
VIP address
Database to Instance One-to-Many Relationship
Prior to Oracle RAC, the terms Oracle instance and Oracle database
were often used interchangeably.
In the RAC world, the distinction is much clearer, because the
instance and database no longer have a one-to-one relationship.
Oracle RAC systems have a one-to-many relationship between the
database and the instances.
Each instance runs on a separate server.
Oracle RAC instances have their own initialization parameter
file, redo thread, redo log files and undo tablespace.
An Oracle RAC database consists of 2+ database instances.
Each instance has its own memory structures and background
processes similar to a non-RAC instance.
Oracle RAC instances also have additional memory structures and
background processes.
According to Oracle’s documentation, an Oracle 11g RAC database can
have as many as one hundred instances.
Implementing a production of a one-hundred-instance RAC system
as one’s first RAC experience should only be done with one’s resume
prominently posted on Monster and HotJobs.com.
In other words, one’s first RAC implementation should not be a
bite that is more than one can chew.
Maximum Availability
For maximum availability, Oracle Data Guard
with Oracle RAC can provide HA even when the local data center is
completely down. The key features of Oracle RAC are high availability
and scalability. Multiple
instances on multiple servers ensure that a single server is not a
single point of failure.
