Performance Tuning :: Oracle Buffer Versus AIX Filesystem Cache
May 7, 2011
I am currently in the favorable situation in which I have excess amounts of memory available on the database server - a single node setup. The server only serves the single instance and no other processing. Database size is around 2.3tb and memory is 50gb. For the majority of processing, AIX is allocating a significant amount (anywhere from 30-40%) of the memory to the AIX file system cache (persistent pages).
I've been trying to find documentation about this, but have not had any luck yet. My guess is that it would be better to allow Oracle to cache this data - meaning increase the SGA target and max size to allow for a larger buffer cache. However, the nice thing about the AIX cache is if process memory is needed, the file system cache gives up pages. If the memory was allocated to the SGA, its pretty much locked in.
I have read several articles stating that a larger buffer cache is not always better, as a larger cache takes more management. But having both of the caches active seem to be a waste of memory, effectively storing the data twice - once in AIX persistent pages and a second time in Oracle database buffer cache.
Considering the below factors, I am planning to increase the buffer cache value from 256Mb to 512Mb.
1. Buffer cache hit ratio value is around 35% even in the normal period. 2. free buffer requested value is below during peak & normal hours below.
Statistic Total per Second per Trans --------------------------------- ------------------ -------------- ------------ free buffer requested 54,694,995 15,226.9 2,523.7 free buffer requested 23,412,674 6,501.7 2,585.9
3. most of the top 5 physical reads & logical reads queries are well tuned and some of queries are doing FTS on small tables (table count min 1500 max 35000). SO indexing option is not required for these queires. But these queries getting executed frequently.
SQL> show sga
Total System Global Area 2148534928 bytes Fixed Size 731792 bytes Variable Size 1879048192 bytes Database Buffers 268435456 bytes Redo Buffers 319488 bytes
5.top 5 waitevents during db slow performance & high cpu utilization (>80%) issue.
Top 5 Timed Events ~~~~~~~~~~~~~~~~~~ % Total Event Waits Time (s) Ela Time -------------------------------------------- ------------ ----------- -------- latch free 1,848,898 153,793 52.00 buffer busy waits 395,280 87,201 29.49 db file scattered read 3,488,648 34,199 11.56 enqueue 4,052 10,897 3.68 CPU time 5,567 1.88
6. Top 5 waitvents during normal activities and CPU utilization is around 40%.
Top 5 Timed Events ~~~~~~~~~~~~~~~~~~ % Total Event Waits Time (s) Ela Time -------------------------------------------- ------------ ----------- -------- CPU time 1,860 45.32 db file scattered read 1,133,669 985 23.99 imm op 776 605 14.73 sbtinfo2 208 139 3.40 sbtbackup 2 123 3.00
1) Is shutting down the DB flush all the data buffers, from the buffer Cache? 2) In any oracle version, do we have any way to flush only the buffer Cache.
I have some confusion about Keep Pool in Buffer Cache.
1. What is the reasoning for placing a table in the KEEP buffer pool because if it is frequently accessed, it will be around when needed (ie if it is constantly being accessed it will not age out) . 2. Would the table be still in the Default Pool if the Keep Pool is not sized and the command is being issued alter TABLE SCOTT.EMP storage (buffer_pool keep) ? 3. If the database is restarted will the table be wiped out of the Keep Pool and again be pinned to the Keep Pool ?
Is there any relationship b/w tuning BUFFER CACHE and BUFFER BUSY WAITS?
1) Buffer Busy Waits are happening as the User process found the same Datablock is being used by another user in the BUFFER CACHE. 2) And also happens, when the server process found the same Datablock are being used in the Datafile.
Does cache buffer chain latch and buffer busy wait event are related to one any another.
Latch definition from Google says : Latches are simple, low-level serialization mechanisms to protect shared data structures in the system global area (SGA).
what does it mean my protect. Does this mean protects from aging as per LRU algorithm and getting removed from SGA or protect from other processes ,say from example from simultaneously DML operations. or both
Does buffer busy wait event occurs , because of the cache buffer chain latch ?
I have a serious doubt in oracle architecture functionality, when a user issues a update statement the data blocks are carried to db buffer cache and where does the changes to the data blocks are made???? Does a copy of the data block is kept in db buffer cache and the changes are made to the block in buffer cache?? or the a copy of the data block is kept in undo tablespace and changes are made to the blocks in the undo tablespace???
In simple the changes to the data blocks are made at db buffer cache or undo tablespace?
I noticed oracle background process ora_fbda_padwsdpr is suffering from buffer busy wait. When i further finding the object, it was on SYS_FBA_FA tables.
what is this is causing BUFFER BUSY WAIT. Also to add we have disabled flashback database.
In order to improve the performance of our live server, I am trying to do an exhaustive comparison with our test environment which is quite quick in spite of the fact that we port the data from Live every month.
There are no obviously slow queries appearing in the the top SQLs of AWR, we have optimised such things already. Right now it is about general uplift rather than SQL based tuning.
I picked up random SQLs and I noticed a marked differences in the execution time. Typically they are 3 to 4 times and there are cases much more than that.
1. I observed that, while the explain plan of the queries are same, trace of the queries give a different picture. I have observed that the recursive calls, consistent gets and sorts(memory) are quite high on Live. 2. I have no solid reasons to say this but my instincts tell me that the recursive calls is the major contributing factor. It is sometimes 2000+ for an SQL. 3. On googling more on that, it finally made me compare the data dictionary on the AWR report of test and Live.
The dc_objects caught my eyes. In that 4 hour AWR, there were about 10 million get requests and the pct miss was ~10. For similar load, the test server had 5 million gets with 0.08 PCT miss for 4 hours.
If you have 3 tables (yr09, yr10,yr11) one with 2009 data, 2010 and 2011 data respectively. And a view (vw_yr091011) with a "union all" on all three.
Question: Will the performance be same for the following two queries ?
Question: Will Oracle read all 3 tables in the view when we search for only one year ?
select count(*) from yr09 where year = 2009;
-- vs
select count(*) from vw_yr091011 where year = 2009;
The following link says yes, the performance remains the same.
Link: [URL]..........
when I tried on a volume of 14000 records. The count came out same but the view took 50 more sec. And the explain plan shows it accessed all three tables.
I'm trying to demonstrate the working of the OCI client result cache. I've set some parameters,orcl> sho parameter result_cache
NAME TYPE VALUE ------------------------------------ ----------- ------------------------------ client_result_cache_lag big integer 3000 client_result_cache_size big integer 1000000 result_cache_max_result integer 5 result_cache_max_size big integer 1984K result_cache_mode string FORCE result_cache_remote_expiration integer 0
If I understand the docs correctly, that should be all that is needed.I've complied and run the cdemoqc and cdemoqc2 OCI demos, but I never get anything in the V$CLIENT_RESULT_CACHE_STATS or CLIENT_RESULT_CACHE_STATS$ views. This is probably because the sessions exit after running the queries.
So I've also tried repeating arbitrary queries through cdemo2, which gives a persistent session, but there is still nothing in those views and furthermore the v$result_cache_objects.scan_count for my queries keeps increasing. So I don't think think the client side cache is working.
i am testing a proc after tuning it but the problem is, it is taking a very very less time which it shouldn't. I know that it is because of the buffer cache and the shared pool. that why i need to clean the cache to retest it.
I cannot bounce the database as other schemas are part of it. so is there any way to clean the cache for that particular schema i.e bouncing any particular schema(i know that the term is not appropriate).
I have two design alternatives and need to understand how expensive (speed) is one of them against the other for a medium size table (100K-200K records):
create table xyz ( f1 number not null, f2 varchar2(20) not null, f3 number not null, f4 varchar2(50),
[code]....
the idea is to optimize the design by using a PK instead of the 3 keys and there is a debate that searching a unique index field(2nd scenario) is of the same speed than searching a PK field (1st scenario).
What's the difference between a dirty buffer and a redo buffer?
My understanding is that a dirty buffer is a changed buffer or whenever data changes in the buffer cache, it's marked as dirty. Also, a redo buffer keeps track of changes that were made to the data, so it's also referring to changed data as well...DWBn writes dirty buffers to disk and LGWR writes redo data to redo log filesHow can we differentiate between the two?
I have a huge table (about 60 gb) partition over range. The index on this table is global index created on 4 columns together. I have a query which is running very slowly. The explain plan is showing the use of this global index.Explain plan is not showing pstart and pend because the index is global.
1)If i issue a DELETE statement to delete a row, will this statement drag any data from the datafile to database buffer? How is the change made by a DELETE statement recorded in buffer cache? How is this change then applied to the data in datafiles after commit?
If sga Buffer Cache Size consume full of SGA_target size, if possible that it will cause performance any issue. I am facing CPU 100% consuming while single query execute, Which query generate monthly report data.
I have two question
1)How to fix the CPU 100% consuming
2)How to find total number user hit oracle specific schema.
Oracle 10.2.0.5 Standard Sga_target : 14G Sga_max :20G Pga :3G
Below SGA details NAME BYTES/1024/1024 -------------------------------- --------------- Fixed SGA Size 2.01795197 Redo Buffers 13.9804688 Buffer Cache Size 13632 Shared Pool Size 640 Large Pool Size 16 Java Pool Size 16 Streams Pool Size 16 Granule Size 16 Maximum SGA Size 20480 Startup overhead in Shared Pool 208 Free SGA Memory Available 6144
Say Database Buffer Cache configured as 2M and my updates may use 4m size,will it throw an error message or update will happen perfectly without any issues?
