Вы находитесь на странице: 1из 5

Oracle Memory Architecture

Oracle uses three kinds of memory structures

(System Global is a large part of memory that all the oracle background processes access.
(Process Global This is memory that is private to a single process or thread and is not accessible by any other process or thread
(User Global

This is memory that is assoicated with your session, it can be found in the PGA or SGA depending on whether
you are connected to the database via shared server
Shared Server - the UGA will be in the SGA
Dedicated Server - the UGA will be in the PGA

There are five memory structures that make up the System Global Area (SGA). The
SGA will store many internal data structures that all processes need access to, cache
data from disk, cache redo data before writing to disk, hold parsed SQL plans and so

The shared pool consists of the following areas:

Shared Pool

Library cache includes the shared SQL area, private SQL areas, PL/SQL procedures and packages the
control structures such as locks and library cache handles. Oracle code is first parsed, then executed ,
this parsed code is stored in the library cache, oracle first checks the library cache to see if there is an
already parsed and ready to execute form of the statement in there, if there is this will reduce CPU time
considerably, this is called a soft parse, If Oracle has to parse it then this is called a hard parse. If there
is not enough room in the cache oracle will remove older parsed code, obviously it is better to keep as
much parsed code in the library cache as possible. Keep an eye on missed cache hits which is an
indication that a lot of hard parsing is going on.
Dictionary cache is a collection of database tables and views containing information about the
database, its structures, privileges and users. When statements are issued oracle will check permissions,
access, etc and will obtain this information from its dictionary cache, if the information is not in the
cache then it has to be read in from the disk and placed in to the cache. The more information held in
the cache the less oracle has to access the slow disks.

The parameter SHARED_POOL_SIZE is used to determine the size of the shared pool, there is no way to
adjust the caches independently, you can only adjust the shared pool size.
The shared pool uses a LRU (least recently used) list to maintain what is held in the buffer, see buffer
cache for more details on the LRU.
You can clear down the shared pool area by using the following command
alter system flush shared_pool;

This area holds copies of read data blocks from the datafiles. The buffers in the cache contain two lists,
the write list and the least used list (LRU). The write list holds dirty buffers which contain modified data
not yet written to disk.
The LRU list has the following

free buffers hold no useful data and can be reused

pinned buffers actively being used by user sessions
dirty buffers contain data that has been read from disk and modified but hasn't been written to

It's the database writers job to make sure that they are enough free buffers available to users session, if
not then it will write out dirty buffers to disk to free up the cache.
There are 3 buffer caches

Buffer cache

Default buffer cache, which is everything not assigned to the keep or recycle buffer pools,
Keep buffer cache which keeps the data in memory (goal is to keep warm/hot blocks in the pool
for as long as possible), DB_KEEP_CACHE_SIZE.
Recycle buffer cache which removes data immediately from the cache after use (goal here is to
age out a blocks as soon as it is no longer needed), DB_RECYCLE_CACHE_SIZE.

The standard block size is determined by the DB_CACHE_SIZE, if tablespaces are created with a different
block sizes then you must also create an entry to match that block size.

DB_2K_CACHE_SIZE (used with tablespace block size of 2k)

DB_4K_CACHE_SIZE (used with tablespace block size of 4k)
DB_8K_CACHE_SIZE (used with tablespace block size of 8k)
DB_16K_CACHE_SIZE (used with tablespace block size of 16k)
DB_32K_CACHE_SIZE (used with tablespace block size of 32k)
buffer cache hit ratio is used to determine if the buffer cache is sized correctly, the higher the value the
more is being read from the cache.
hit rate = (1 - (physical reads / logical reads)) * 100

You can clear down the buffer pool area by using the following command
alter system flush buffer_cache;

The redo buffer is where data that needs to be written to the online redo logs will be cached
temporarily before it is written to disk, this area is normally less than a couple of megabytes in size.
These entries contain necessary information to reconstruct/redo changes by the INSERT, UPDATE,
The contents of this buffer are flushed:
Redo buffer

Every three seconds

Whenever someone commits a transaction
When its gets one third full or contains 1MB of cached redo log data.
When LGWR is asked to switch logs

Use the parameter LOG_BUFFER parameter to adjust but be-careful increasing it too large as it will
reduce your I/O but commits will take longer.
This is an optional memory area that provide large areas of memory for:

Large Pool

Shared Server - to allocate the UGA region in the SGA

Parallel execution of statements - to allow for the allocation of inter-processing message
buffers, used to coordinate the parallel query servers.
Backup - for RMAN disk I/O buffers

The large pool is basically a non-cached version of the shared pool.

Use the parameter LARGE_POOL_SIZE parameter to adjust
Used to execute java code within the database.
Java Pool
Use the parameter JAVA_POOL_SIZE parameter to adjust (default is 20MB)
Streams are used for enabling data sharing between databases or application environment.
Streams Pool
Use the parameter STREAMS_POOL_SIZE parameter to adjust

The fixed SGA contains a set of variables that point to the other components of the
SGA, and variables that contain the values of various parameters., the area is a kind of
bootstrap section of the SGA, something that Oracle uses to find other bits and pieces
of the SGA
For more information regarding setting up the SGA click here.
The PGA (Process Global Area) is a specific piece of memory that is associated with
a single process or thread, it is not accessible by any other process or thread, note that
each of Oracles background processes have a PGA area. The UGA (User Global
Area) is your state information, this area of memory will be accessed by your current

session, depending on the connection type (shared server) the UGA can be located in
the SGA which is accessible by any one of the shared server processes, because a
dedicated connection does not use shared servers the memory will be located in the

Shared server - UGA will be part of the SGA

Dedicated server - UGA will be the PGA

Memory Area

Dedicated Server

Shared Server

Nature of session memory



Location of the persistent area



Location of part of the runtim area for select




Location of the runtime area for DML/DDL




Oracle creates a PGA area for each users session, this area holds data and control
information, the PGA is exclusively used by the users session. Users cursors, sort
operations are all stored in the PGA. The PGA is split in to two areas

Session Information
(runtime area)
Stack space
(private sql area)

PGA in an instance running with a shared server requires additional memory

for the user's session, such as private SQL areas and other information.
The memory allocated to hold a sessions variables, arrays, etc and other
information relating to the session.

Automatic PGA Management

To reduce response times sorts should be performed in the PGA cache area (optimal
mode operation), otherwise the sort will spill on to the disk (single-pass / multiplepass operation) this will reduce performance, so there is a direct relationship between
the size of the PGA and query performance. You can manually tune the below to
increase performance

sort_area_size - total memory that will be used to sort information before

swapping to disk

sort_area_retained_size - memory that is used to retained data after a sort

hash_area_size - memory that will would be used to store hash tables
bitmap_merge_area_size - memory Oracle uses to merge bitmaps retrieved
from a range scan of the index.

Staring with Oracle 9i there is a new to manage the above settings that is to let oracle
manage the PGA area automatically by setting the parameter following parameters
Oracle will automatically adjust the PGA area basic on users demand.

workarea_size_policy - you can set this option to manual or auto (default)

pga_aggregate_target - controls how much to allocate the PGA in total

Oracle will try and keep the PGA under the target value, but if you exceed this value
Oracle will perform multi-pass operations (disk operations).

System Parameters

manual or auto (default)


total amount of memory allocated to the PGA

PGA/UGA amount used

select a.name, to_char(b.value, '999,999,999') value

from v$statname a, v$mystat b
where a.statistic# = b.statistic#
and a.name like '%ga memory%';

Display if using memory or disk sorts

set autotrace traceonly statistics;

set autotrace off;

Display background process PGA memory


select program, pga_used_mem, pga_alloc_mem, pga_max_mem from