|
Previous versions
of Oracle OPS were not very good for OLTP systems, due to pinging and
performance issues. Not only is performance improved with RAC, but also
reliability. In order to lose the DB with an Oracle9i RAC you would have to
either lose the entire LAN, lose the SAN drives, or lose all servers that
are running RAC. With modern built in redundancies for the network, SAN
drives, and servers, the scenario of total loss is minimized greatly.
Attributes of an Oracle RAC Cluster
The Oracle9i RAC allows for a single or multi system image. This means that
the root file and other system files can be either localized in internal or
non-shared disks for each server in the RAC or each server can load the
operating system from a central set of disks.If the RAC cluster uses a SAN or other network storage device it is known as
a Shared System Disk (root) system.
A RAC system must use a cluster file system or a raw partition, where any
server can read or write to any disk in the shared disk subsystem. This
allows access to all data files, control files, and redo and rollback areas
by any instance. This ability to access all disks allows for instance
recovery after an instance failure has occurred. All surviving nodes
automatically absorb the failed instance’s tasks until the failed instance
is brought back online, at which time it is fully synchronized and restored
to service automatically.
A RAC cluster provides for automatic shared execution of Oracle
applications. This means that for any Oracle instance application, all
queries and other processing are automatically shared among all of the
servers in the RAC cluster.
The sharing of application processing by all servers in the RAC cluster
leads to automatic load balancing across all cluster members. The ability of
a RAC cluster to provide shared application execution and automatic load
balancing leads to the true scalability of applications without code or data
changes.
One limit on shared-none or federated databases was the amount of time
required for failover. Due to the requirements for rebuilding or rehashing
of data, fail-over time in a non-shared disk environment can be prohibitive.
In a RAC cluster, fail-over time is virtually nil, since Oracle handles the
redistribution of work automatically. Furthermore, the rehashing of data is
not required because all participating nodes in the RAC cluster share the
same data files.
The number of servers that your LAN architecture allows you to connect into
the disk farm is the only limit to the number of nodes in a RAC cluster. In
the SQL Server 2000, the cluster database has a hard limit of 2 or 4
servers, depending on the level of the server license for Windows 2000 that
is loaded on the server. The convoy effect limits the number of servers that
an IBM type structure allows.
Due to the cache fusion and the high-speed cluster interconnect, the
Oracle9i RAC allows for a virtually unlimited number of nodes in a RAC
cluster. The maximum number of instances that can be interconnected is OS
platform-dependent. This indicates that you need to plan ahead and determine
just how large the RAC cluster is anticipated to grow, before making
operating system decisions concerning your RAC environment.
Oracle9i RAC allows for the following protocols to be used in the cluster
interconnect:
The speed of the cluster interconnect should be, at a minimum, 100 mbit/sec.
The distance allowed between nodes on a cluster interconnect is system and
LAN-specific. Again, plan ahead. Some forms of interconnect allow only a
couple hundred yards, some allow miles; your choice for network protocol
will determine whether nodes must reside in the same room, building, or be
spread across your corporate campus.
Oracle RAC offers greater disaster tolerance than either UDB/DB2 or
Microsoft SQLServer, because if single or multiple nodes fail, the load is
not redistributed between all of the remaining nodes. If the database is
globally distributed then a disaster to a single site will not affect other
sites.
Building an Oracle RAC Cluster
The first step in building a RAC cluster is to establish the connecting LAN
to a SAN or other network storage device via a HUB, fiber channel switch, or
shared SCSI (for systems not requiring high performance, IEEE1394 or
Firewire drives can also be used). Once the LAN and SAN are configured,
servers are attached and configured. If the root, user, and other systems
are moved into the shared storage and shared among all cluster members, it
greatly reduces the amount of system management overhead and database
overhead.
Once the LAN, SAN, and base servers are configured, the cluster manager
software from Oracle is installed on all nodes that will participate in the
Oracle cluster configuration. Following the cluster manager installation,
the main node is loaded with the Oracle software, which is automatically
copied to all other nodes, as directed by the person doing the installation.
 |
If you like Oracle tuning, check out my latest book
"Oracle
Tuning: The Definitive Reference".
It's 980 pages of hard-core tuning insights, tips and
scripts, and you can buy it direct from the publisher for 30%-off. |
|
|
|
