Shared Cluster Architecture
Unlike the ‘shared nothing’ Google search
platform, Oracle RAC is a shared everything
database.
The following items are shared by all RAC
instances.
Since Oracle RAC is a shared everything system,
the data files, control files, SPFILEs and redo
log files must reside on cluster aware shared
storage.
Storage Options
The shared storage options for Oracle 11g RAC
are ASM, OCFS for Windows, OCFS2 for Linux, NFS
(now with Direct NFS in 11g), certified
third-party cluster-aware volume manager,
certified network file system, and raw devices.
Oracle requires that RAC implementations
using Network Attached Storage (NAS) use their
own private network, which is separate from the
interconnect.
Metalink Note 578455.1 indicates Oracle’s
support of raw devices is going away in Oracle
12g.
Virtual IP Address
Unlike a single instance database, an Oracle RAC
instance can be connected to a client through
the server’s IP address and through VIP
addresses.
The VIP address or corresponding virtual
host name is the logical way to connect to the
RAC database, because of the built-in failover
ability.
Interconnect
The interconnect is a critical component of the
Oracle RAC architecture.
From Oracle:
“The interconnect network is a private
network that connects all of the servers in the
cluster. The interconnect network uses a switch
(or multiple switches) that only the nodes in
the cluster can access. Configure User Datagram
Protocol
(UDP) on a Gigabit you can configure Oracle
Clusterware
to use either the UDP or Reliable Data Socket
(RDS) protocols. Windows clusters use the TCP
protocol. Crossover cables are not supported for
use with Oracle Clusterware interconnects.”
Using 10 Gigabit Ethernet
as the interconnect is not mentioned above by
Oracle, but is a good choice because it can move
1,250 MB/s versus Gigabit Ethernet’s 125 MB/s.
Cache Fusion
Cache Fusion
is a core part of Oracle RAC.
The System Global Area
(SGA) consists of multiple components, one of
which is the buffer cache
.
In RAC, cache fusion combines the buffer cache,
allowing the buffer cache size to be much
larger.
Cache fusion
traffic and other inter-instance communication
takes place over the RAC interconnect private
network.
A data block initially resides on disk.
As queries are run, data blocks are
copied from disk to the requesting instance’s
buffer cache.
If another instance needs any of the data
blocks that are now in the first instance’s
buffer cache, the second instance will obtain
the blocks from the first instance, preventing a
more expensive trip to disk.
The Global Cache Service (GCS) and the
Global Enqueue Service
(GES) processes keep track of the status of each
cached block using a memory resident data
structure called the Global Resource Directory
(GRD).
There are a number of RAC specific processes
which will be explained in more detail in other
chapters.
The processes are ACMS,
GTX0-j, LMON, LMD, LMS, LCK0,
RMSn
and RSMN.