Tuning Global Cache - Expert Oracle RAC Performance Diagnostics and Tuning - page 380

buffer cache of an Oracle RAC instance by a small percentage. Cache Fusion (illustrated in Figure 13-1 ) moves current
and consistent data buffers from an instance that contains the block to other instances. This saves an I/O operation
to re-read the blocks from disk and is faster than a disk access. When a consistent image of the block is needed or a
changed block is required on another instance, Cache Fusion transfers the block image directly between the affected
instances using the private interconnect.

GRD

SGA

GRD

SGA Cache Fusion

LGWR DBWR LGWR

DBWR LMS LMD LMD LMS

Shared storage subsystem

Redologs Database Files Redologs

Figure 13-1. Cache Fusion

The GCS processes on each node manage the cache synchronization by using cluster interconnect. RAC
addresses correct transactional concurrency between instances. The different scenarios of block sharing are broadly
stated as the following:

• Concurrent reads on multiple nodes: This occurs when two or more instances participating
in the clustered configuration are required to read the same data block. The block is shared
between instances via the cluster interconnect. Initially, the first instance that reads the block
from disk and the subsequent instances that require access to the same block will request it
via the cluster interconnect. Access rights and ownership of the block are transferred with
the data.

• Concurrent reads and writes on different nodes: This is a mixture of read/write operations
on data. A data block available on any of the participating instances could be modified by
multiple instances, whereas the version of the data block in the physical database may be an
earlier copy until such time that the DBWR on one of the instances decides to write the block
to database. Transactions where blocks are requested from another instance either for a read
or for a write operation use the interconnect. A block can be read as is, that is, in a current
version, or a read consistent version could be built by applying undo.

• Concurrent writes on different nodes: This is a situation where multiple instances make
requests for modification to the same data block frequently. In these two situations where
modifications are involved, global consistency of data blocks are maintained by assigning
access modes and versioning.

RAC Specific Background Processes
A RAC system comprises two or more nodes (instances) accessing a common shared database (i.e., one database
is mounted and open by multiple instances concurrently). In this case, each instance will have all the background
process used in a stand-alone configuration plus the additional background processes required specifically for RAC.