0% found this document useful (0 votes)
167 views

Tuning SGA

The document discusses tuning the Oracle shared global area (SGA). It covers tuning the shared pool, buffer cache, redo log buffer, and data dictionary cache. The shared pool should be tuned by monitoring library cache hit ratios, invalidations, and UGA usage. The buffer cache can be tuned by increasing the DB_BLOCK_BUFFERS parameter if the cache hit ratio is below 90%. The redo log buffer size should be increased if the redo log space request ratio is greater than 1 in 5000.

Uploaded by

Suresh Patnaik
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
167 views

Tuning SGA

The document discusses tuning the Oracle shared global area (SGA). It covers tuning the shared pool, buffer cache, redo log buffer, and data dictionary cache. The shared pool should be tuned by monitoring library cache hit ratios, invalidations, and UGA usage. The buffer cache can be tuned by increasing the DB_BLOCK_BUFFERS parameter if the cache hit ratio is below 90%. The redo log buffer size should be increased if the redo log space request ratio is greater than 1 in 5000.

Uploaded by

Suresh Patnaik
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 7

Oracle Tuning of SGA

1.0 Tuning the shared pool

1. Shared Pool includes:

a. library cache - holds PL/SQL blocks ( Procedures etc..)

b. data dictionary cache - database dictionary objects

c. user global area -

2. init.ora parameters -

a. SHARED_POOL_SIZE. To determine proper size, set size to a very large


amount and then run application. Sum the totals for the following queries:

i. select sum(shareable_mem) from v$db_object_cache where type = ‘PACK-


AGE’ or type = ‘PACKAGE BODY’ or type = ‘FUNCTION’ or type = ‘PROCEDURE’;

ii. select sum(shareable_mem) from v$sqlarea where executions > 5;

iii. select sum(250 * user_opening) from v$sqlarea; during peak application use

or in a test environment

select 250 * value bytes_per_user from v$sesstat s, v$statname n where s.sta-


tistic# = n.statistic# and n.name = ‘opened cursors current’ and s.sid = 15;

iv. Allow a small allowance for dynamic sql

b. SHARED_POOL_SIZE is too small when request_failures > 0 and increasing

3. Tuning the Library Cache

a. Reduce library cache misses by allocating enough memory

b. Ensure SQL statements use generic code and bind variables.

i. V$LIBRARYCACHE - check GETHITRATIO (GETHITS/GETS) > 90%

Oracle Tuning of SGA January 23, 2004 1


ii. V$LIBRARYCACHE - Reloads < (0.01)PINS. If greater than 1% increase
SHARED_POOL_SIZE.

c. Avoid changes to schema object referenced by a sql statement. The resulting


invalidation requires a reload.

i. V$LIBRARYCACHE - check invalidations column

d. Avoid fragmentation

i. Use reserved space for large memory requirements

- initially set SHARED_POOL_RESERVED_SIZE to 10% of


SHARED_POOL_SIZE.

- V$SHARED_POOL_RESERVED

-- goal of request_misses = 0

-- last_failure_size > shared_pool_reserved_min_alloc

-- avg_free_size > shared_pool_reserved_min_alloc

- If SHARED_POOL_RESERVED_SIZE is too small

--increase SHARED_POOL_SIZE and


SHARED_POOL_RESERVED_SIZE

- if SHARED_POOL_RESERVED_SIZE is too large (free_space = >


50% of shared_pool_reserved_size)

ii. Pin often-used large objects

- objects include compiled triggers, sequences, large procedural objects


such as STANDARD and DIUTIL.

- run dbmspool.sql to create package DBMS_SHARED_POOL and the


KEEP and UNKEEP procedures.

iii. Use small PL/SQL packaged functions instead of large anonoymous blocks

- run script select sql_text from v$sqlarea where command_type = 47


and length(sql_text) > 500;

e. Reduce UGA memory usage.

Oracle Tuning of SGA January 23, 2004 2


4. Tuning Data Dictionary Cache

i. V$ROWCACHE: ratio of GETMISSES/GETS < 15% else increase


SHARED_POOLSIZE. The column GETMISSES cannot have a zero value due to load-
ing at startup.

2.0 Tuning the Buffer Cache

1. Buffer Cache: block can be either free buffers, dirty blocks, pinned buffers. Managed by
DBWR process.

2. init.ora parameter DB_BLOCK_BUFFERS

3. Tuning buffer cache

a. cache hit ratio > 90%

i. calculate hit ratio using V$SYSSTAT

- Hit Ratio = 1- (physical reads/ (db block gets + consistent gets))

b. Use utilities to estimate the hit-ratio change when buffers are added or removed.

i. If hit ratio is low test impact of adding buffers by:

- set init.ora DB_BLOCK_LRU_EXTENDED_STATISTICS = ???

- start database for normal use.

- After a period of normal running query table V$RECENT_BUCKET (count)

ii. Use the following queries

- select sum(count) ach from v$recent_bucket where rownum < 20; Use ach
value in statement below.

- select 1- ((phy.value -ach) / (cur.value + con.value )) “CACHE HIT RATIO”


from v$sysstat cur, v$sysstat con, v$sysstat phy where cur.name = ‘db block gets’ and
con.name =’consistent gets’ and phy.name =’physical reads’;

iii. If hit ratio is high and reducing memory usage is desired then:

- set ini.ora DB_BLOCK_LRU_STATISTICS = TRUE and reset when done.

- start database for normal use

Oracle Tuning of SGA January 23, 2004 3


- After a period of normal running query table
V$CURRENT_BUCKET(count).

iv. Use the following queries:

- select sum(count) acm from v$current_bucket where indx


>=(db_block_buffers) - 20;

- select 1- ((phy.value -ach) / (cur.value + con.value )) * 100 “CACHE HIT


RATIO” from v$sysstat cur, v$sysstat con, v$sysstat phy where cur.name = ‘db block gets’
and con.name =’consistent gets’ and phy.name =’physical reads’;

c. Rules for increasing DB_BLOCK_BUFFERS

i. cache hit ratio is less than 90%

ii. There is added memory for other processes as determined by amount of page
faults.

iii. Previous increase of DB_BLOCK_BUFFERS worked.

4. Using Multiple Buffer Pools

a. DBA can define 1,2, or 3 pools. (DEFAULT, KEEP, RECYCLE). ‘DEFAULT’ is


DB_BLOCK_BUFFERS. KEEP buffer pool is for objects(especially large) that are fre-
quently used but may age out of the default pool. RECYCLE buffer pool is for objects
that are only need for a given transaction.

i. BUFFER_POOL_KEEP = (BUFFERS:1200, LRU_LATCHES:1)

ii. BUFFER_POOL_RECYCLE = (BUFFERS:1200, LRU_LATCHES:2)

b. Pool blocks taken from DB_BLOCK_BUFFERS. Total cannot exceed


DB_BLOCK_BUFFERS

c. For 1 latch at least 50 blocks. Latches taken from DB_BLOCK_LRU_LATCHES.


Total defined latches for each pool cannot exceed DB_BLOCK_LRU_LATCHES.

d. Sizing

i. Use analyze...estimate statistics. ex. Analyze Table atable Estimate Statistics.

ii. Select table_name, blocks from dba_tables where owner = ‘??’ and
table_name=’aname’; Sum the blocks from DBA_TABLES, DBA_INDEXES,
DBA_CLUSTERS for each object.

iii.

Oracle Tuning of SGA January 23, 2004 4


3.0 Tuning the Redo Log Buffer
1. Redo Log Buffer - usually a small area not more than 1M. Default value usually 4*max
block size which is OS specific. Size must be a multiple of system block size.

2. init.ora - LOG_BUFFER = ???? in bytes

3. Tuning guidelines:

a. V$SYSSTAT - redo log space requests to the number of redo entries should not be
greater than 1 in 5000.

i. select (req.value*5000)/entries.value “Ratio”

from v$sysstat req, v$sysstat entries

where req.name = ‘redo log space requests’

and entries.name = ‘redo entries’;

ii. increase LOG_BUFFER size until ratio is stable.

b. Check for cause of LGWR slowness

i. Redo log file should be on separate drives

ii. Waits for log switch to occur

- V$SESSION_WAIT - see if there are any waits for log switch. The
SECONDS_IN_WAIT value of the ‘log buffer space’ event indicates time spent waiting
for space in the redo log buffer because log switch has not occured.

-- This indicates that the buffers are being filled up faster then LGWR is writing.

-- This might also mean disk I/O contention on the redo log files.

-- select sid, event, seconds_in_wait, state from v$session_wait

where event like ‘log%’;

iii. Log file switch waits because of log switches

- V$SYSTEM_EVENT : ‘log file switch completion’

-- select event, total_waits, time_waited, average_wait from v$system_event

where event like ‘log file switch completion%’;

-- Increase the size of the redo log files and or add groups.

Oracle Tuning of SGA January 23, 2004 5


iv. LGWR is waiting on DBWR

- grep alert.log for “CHECKPOINT NOT COMPLETE”

- V$SYSTEM_EVENT - check ‘log file switch(checkpoint incomplete)’

-- select event, total_waits, time_waited, average_wait from v$system_event

where event like ‘log file switch (check%’;

-Adjust LOG_CHECKPOINT_INTERVAL, LOG_CHECKPOINT_TIMEOUT

- Adjust size and number of redo log groups

v. LGWR is waiting on the archiver.

- V$SYSTEM_EVENT - check ‘log file switch(archiving needed)’

-- select event, total_waits, time_waited, average_wait from v$system_event

where event like ‘log file switch (arch%’;

- check that archive device is not full

- add redo log groups

- regulate archiving speed: LOG_ARCHIVE_BUFFERS,


LOG_ARCHIVE_BUFFER_SIZE.

- check if LOG_BLOCKCHECKSUM=TRUE. This adds overhead.

Oracle Tuning of SGA January 23, 2004 6


January 23, 2004 7

You might also like