2. New view in new ST04 under Peformance > Stats Info > SGA

3. Dynamic SGA
- allows resezing SGA components ( buffer, shared pool and large pool)  without restarting at runtime. The others SGA components need to be restarted.
- sga_max_size = maximum size for total SGA. If not defined, the value is the total sga size.
- large sga will cause swapping.

4. SGA info
- V$SGA_CURRENT_RESIZE_OPS  = Current SGA size change operation(s)
- V$SGA_RESIZE_OPS = List of the last 100 SGA size change operations carried out
- V$SGA_DYNAMIC_COMPONENTS = Display detail info and history
- V$SGA_DYNAMIC_FREE_MEMORY = Display of the memory available for future SGA enhancements

5. Tuning DB buffer
- Cache size advice based on data buffer quality/hit ration and V$DB_CACHE_ADVISE (parameter DB_CACHE_ADVICE = ON or STATISTIC_LEVEL = typical ).
- SQL>ALTER SYSTEM SET STATISTICS_LEVEL=’TYPICAL’;
- high hit ration doesnt mean good sizing, as it could be due to expensive SQL. if ration buffer get : user call  > 15, hit ratio is   not a good indicator. Check for log read/user call value.ST04 and ST04OLD use diff method to collection of hit ratio.
- use ST04, cache advisory tab, the check the values of physical read for a range of cache size. Objective is to have less phy     read or until the phy read is constant after adding more cache size.

http://download.oracle.com/docs/cd/B10500_01/server.920/a96533/memory.htm

6. Tuning Shared pool
- If the shared pool is too small, then extra resources are used to manage the limited amount of available space.  This consumes CPU and latching resources, and causes contention and possibly ORA-4031
- Use v$shared_pool_advice
- Check the freepool in v$sgastat to see free memory

Tuning PGA
1. Min PGA aggr value:
SELECT round(PGA_TARGET_FOR_ESTIMATE/1024/1024) target_mb,
ESTD_PGA_CACHE_HIT_PERCENTAGE cache_hit_perc,
ESTD_OVERALLOC_COUNT
FROM v$pga_target_advice;

2. v$pga_target_advice; = ESTD_OVERALLOC_COUNT must be zero

3. multipass, onepass or optimal can be monitored from ST04 PGA monitor- status – workarea executions.

Note:
- Both em and heap uses swap space, ensure swap file is increased correspondingly.
- For unix ensure shm pools (e.g HP-ux pool 10 and 40) are enough.
- Note 425207 SAP memory management, current parameter ranges

2. ST02 storage report
- Storage shared between wp = total usage for sap buffers. they take hp-ux pool.
- user storage for each wp = mem of each wp x total number of wp
- size of ext = em/initial_size_MB set.
- virtual mem allocated = mem req for sap.

3. Allocation sequence
- DIA = roll area > em > heap
- NONDIA = roll area >heap > em
- Use RSMEMORY to redefine seq. (0 – Roll area, 1 – Ext Memory, 2 -  Heap)
- SAP Note 177226 Documentation rsmemory

Resource
1. http://www.oracle.com/technology/obe/2day_dba/gettingstarted/gettingstarted.htm
2. 355770 – Oracle Enterprise Manager Database Control
3. http://technopark02.blogspot.com/2008/02/oracle-10g-setting-up-oracle-enterprise.html

a. 540463 – FAQ: Consistency Checks + Block Corruptions
b. 23345 – Consistency check of ORACLE database

1. Analyze table validate structure cascade
Runtime: long – locks the table examined currently against change accesses
Only feasible in times of minimal work load or – even better – when the SAP system has been stopped
starting with Oracle 9 the extension ‘analyze table validate structure cascade ONLINE’; does not lock the table anymore checks tables and indexes if an analyze (online or not online) is running on a table and you try to alter storage parameters, parallel degree etc from this table then all further sessions accesing the table with write or read access are locked until the analyze is finished. Further details can be found in note 619188. reads the blocks to be checked in the SGA, causing possible temporary deterioration in the buffer quality of the DB block buffer does NOT check data stored in lob columns

2.Export
Runtime: medium – feasible in a running system. Performance loss as a result of the export processes. If you access extensive tables in write mode during export, there is a danger of ORA-1555 “snapshot too old”. However, you can prevent this by sufficiently dimensioned rollback segments only checks tables certain kinds of corruptions are exported without an error occuring. If you later try to reorganize this type of table (export/import), you will have problems during import. reads the blocks to be checked into the SGA which is why the buffer quality of the DB block buffer is adversely affected for a short while.

3. dbverify
Runtime: short – feasible in running operation (UNIX: all oracle releases; Windows Oracle9 or newer)
not only checks tables and indexes but also blank DB blocks does not check cross references between tables and indexes . Reads the blocks WITHOUT loading them in the SGA, which is why the buffer quality of the DB block buffer is not affected the only procedure that is also possible on data files restored from a backup (without these files having to belong to a DB) does together with blockchecksums according note 923919 an extended check on blocks and can then find special kinds of cross reference corruptions because those may cause wrong checksums. Wrong cross references caused by lost writes do NOT lead to wrong checksums and therefore cannot be detected.
Also cross reference failures already occuring in main memory BEFORE the checksum for the table and index blocks are calculated cannot be detected because the checksums base already on bad data.
Anyway dbverify together with blockchecksums are a considerable alternative if an analyze table validate structure cascade on the complete DB lasts to long.

2.  The PGA allocation is dynamic and can be affected by the parameters SORT_AREA_SIZE, HASH_AREA_SIZE,
BITMAP_MERGE_AREA_SIZE and CREATE_BITMAP_AREA_SIZE or PGA_AGGREGATE_TARGET when you use the automatic PGA administration

3. V$DB_CACHE_ADVICE – size advisor

4. SGA_MAX_SIZE – dynamic SGA. Key SGA components can be altered dynamically – buffer cache, shared pool and large pool.

The reason why big heaps (4g for example) are not recommended is that the bigger the heap the longer it takes for the garbage collector to run through it. So there is a trade off between less gc-s in the beginning when the JVM is started and much longer ones, when the heap is for example half full. Therefore, larger max heap doesnt related to better performance. If you have huge resource, you add more server nodes you can scale more easily and serve a larger number of users.

a. 723909 – Java VM settings for J2EE 6.40/7.0
b. 876722 – Java heap size doesn’t update or appears twice in cofig tool