physical reads...

Similar Questions

• "Physics read total multi block applications ' wrong?

  Hello

  The statistics "block the physical demands reading total multi" should count the number of calls to e/s over a reading block (doc here). So I expect to be equal to the 'file db scattered read' (or 'direct path read' if read series direct-path of access is used).

  But this isn't what I get in the following unit test:

  SQL > create table DEMO pctfree 99 as select rpad('x',1000,'x') n from dual connect by level < = 1000;

  -I get the v$ filestat statistics in order to make a difference at the end:

  SQL > select phyrds, phyblkrd, singleblkrds from v$ filestat when file #= 6;

  PHYRDS PHYBLKRD SINGLEBLKRDS

  ---------- ---------- ------------

  6291-96488-3286

  I'm doing a table scan complete on a table of 1000 blocks, forcing sql_tracing and stamped readings

  SQL > connect demo/demo

  Connected.

  SQL > alter session set "_serial_direct_read" = never;

  Modified session.

  SQL > alter session set events =' waiting sql_trace = true ";

  Modified session.

  SQL > select count (*) in the DEMO.

  COUNT (*)

  ----------

  1000

  SQL > alter session set events = 'off sql_trace. "

  session statistics see the 30 requests for e/s, but only 14 requests for multi block:

  SQL > select name, value of v$ mystat join v$ statname using(statistic#) whose name like "% phy" and the value > 0;

  NAME                                          VALUE

  ---------------------------------------- ----------

  total read physics applications of e/s 30

  physical read total multi block 14 requests

  total number of physical read 8192000 bytes

  physical reads 1000

  physical reads cache 1000

  physical read IO request 30

  bytes read physical 8192000

  physical reads cache prefetch 970

  However, I did 30 'db file squattered read:

  SQL > select event, total_waits from v$ session_evenement where sid = sys_context ('userenv', 'sid');

  TOTAL_WAITS EVENT

  ---------------------------------------- -----------

  File disk IO 1 operations

  1 log file sync

  db file scattered read 30

  SQL * Net message to client 20

  SQL * Net client message 19

  V$ FILESTAT counts them as multiblock reads as follows:

  SQL > Select phyrds-6291 phyrds, phyblkrd-96488 phyblkrd, singleblkrds-3286 singleblkrds in v$ filestat when file #= 6

  PHYRDS PHYBLKRD SINGLEBLKRDS

  ---------- ---------- ------------

  30       1000            0

  And this is confirmed by sql_trace

  SQL > column tracefile new_value by trace file

  SQL > select the process trace file $ v where addr = (select paddr in session $ v where sid = sys_context ('USERENV', 'SID'));

  TRACE FILE

  ------------------------------------------------------------------------

  /U01/app/Oracle/diag/RDBMS/dbvs103/DBVS103/trace/DBVS103_ora_31822.TRC

  SQL > host mv and tracefile series-live - path.trc

  SQL > host grep ^ WAITING series-live - path.trc | grep read | NL

  1. WAIT #139711696129328: nam = "db-scattered files reading" ela = 456 file #= 6 block #= 5723 blocks = 5 obj #= tim 95361 = 48370540177

  2. WAIT #139711696129328: nam = "db-scattered files reading" ela = 397 file #= 6 block #= 5728 blocks = 8 obj #= tim 95361 = 48370542452

  3. wait FOR #139711696129328: nam = "db-scattered files reading" ela = 449 file #= 6 block #= 5737 blocks = 7 obj #= tim 95361 = 48370543216

  4 WAIT #139711696129328: nam = "db-scattered files reading" ela = 472 file #= 6 block #= 5744 blocks = 8 obj #= tim 95361 = 48370543816

  5. WAIT #139711696129328: nam = "db-scattered files reading" ela = 334 file #= 6 block #= 5753 blocks = 7 obj #= tim 95361 = 48370544276

  6 WAIT #139711696129328: nam = "db-scattered files reading" ela = 425 file #= 6 block #= 5888 blocks = 8 obj #= tim 95361 = 48370544848

  7 WAIT #139711696129328: nam = "db-scattered files reading" ela = 304 case #= 6 block #= 5897 blocks = 7 obj #= tim 95361 = 48370545370

  8 WAIT #139711696129328: nam = "db-scattered files reading" ela = 599 file #= 6 block #= 5904 blocks = 8 obj #= tim 95361 = 48370546190

  9 WAIT #139711696129328: nam = "db-scattered files reading" ela = 361 file #= 6 block #= 5913 blocks = 7 obj #= tim 95361 = 48370546682

  10 WAIT #139711696129328: nam = "db-scattered files reading" ela = 407 file #= 6 block #= 5920 blocks = 8 obj #= tim 95361 = 48370547224

  11 WAIT #139711696129328: nam = "db-scattered files reading" ela = 359 file #= 6 block #= 5929 blocks = 7 obj #= tim 95361 = 48370547697

  12 WAIT #139711696129328: nam = "db-scattered files reading" ela = 381 file #= 6 block #= 5936 blocks = 8 obj #= tim 95361 = 48370548287

  13 WAIT #139711696129328: nam = "db-scattered files reading" ela = 362 files #= 6 block #= 6345 blocks = 7 obj #= tim 95361 = 48370548762

  14 WAIT #139711696129328: nam = "db-scattered files reading" ela = 355 file #= 6 block # 6352 blocks = 8 obj #= tim 95361 = 48370549218

  15 WAIT #139711696129328: nam = "db-scattered files reading" ela = 439 file #= 6 block #= 6361 blocks = 7 obj #= tim 95361 = 48370549765

  16 WAIT #139711696129328: nam = "db-scattered files reading" ela = 370 file #= 6 block #= 6368 blocks = 8 obj #= tim 95361 = 48370550276

  17 WAIT #139711696129328: nam = "db-scattered files reading" ela = 1379 file #= 6 block #= 7170 blocks = 66 obj #= tim 95361 = 48370552358

  18 WAIT #139711696129328: nam = "db-scattered files reading" ela = 1205 file #= 6 block #= 7236 blocks = 60 obj #= tim 95361 = 48370554221

  19 WAIT #139711696129328: nam = "db-scattered files reading" ela = 1356 file #= 6 block #= 7298 blocks = 66 obj #= tim 95361 = 48370556081

  20 WAIT #139711696129328: nam = "db-scattered files reading" ela = 1385 file #= 6 block #= 7364 blocks = 60 obj #= tim 95361 = 48370557969

  21 WAIT #139711696129328: nam = "db-scattered files reading" ela = 832 file #= 6 block #= 7426 blocks = 66 obj #= tim 95361 = 48370560016

  22 WAITING #139711696129328: nam = "db-scattered files reading" ela = 1310 file #= 6 block #= 7492 blocks = 60 obj #= tim 95361 = 48370563004

  23 WAIT #139711696129328: nam = "db-scattered files reading" ela = 1315 file #= 6 block # 9602 blocks = 66 obj #= tim 95361 = 48370564728

  24 WAIT #139711696129328: nam = "db-scattered files reading" ela = 420 file #= 6 block # 9668 blocks = 60 obj #= tim 95361 = 48370565786

  25 WAIT #139711696129328: nam = "db-scattered files reading" ela = 1218 file #= 6 block # 9730 blocks = 66 obj #= tim 95361 = 48370568282

  26 WAIT #139711696129328: nam = "db-scattered files reading" ela = 1041 file #= 6 block #= 9796 blocks = 60 obj #= tim 95361 = 48370569809

  27 WAIT #139711696129328: nam = "db-scattered files reading" ela = file No. 300 = 6 block #= 9858 blocks = 66 obj #= tim 95361 = 48370570501

  28 WAIT #139711696129328: nam = "db-scattered files reading" ela = 281 file #= 6 block #= 9924 blocks = 60 obj #= tim 95361 = 48370571248

  29 WAIT #139711696129328: nam = "db-scattered files reading" ela = 305 file #= 6 block #= 9986 blocks = 66 obj #= tim 95361 = 48370572021

  30 WAIT #139711696129328: nam = "db-scattered files reading" ela = 347 file #= 6 block #= 10052 blocks = 60 obj #= tim 95361 = 48370573387

  So, I'm sure that I did 30 reading diluvium but block total multi physical read request = 14

  No idea why?

  Thanks in advance,

  Franck.

  Couldn't leave it alone.

  A ran some additional tests with sizes of different measure, with and without the SAMS.

  It seems that my platform includes no readings of scattered file db less than 128KB in the physical count. (or 16 blocks, given that I was with a block size of 8 KB).

  Concerning

  Jonathan Lewis

• The Oracle trace: number of physical reads of the signifier of disk buffer

  Hi all

  
  Since yesterday, I was under the impression that, in a trace file Oracle, the "number of physical buffer of disk reads" should be reflected in the wait events section.

  Yesterday we add this trace file (Oracle 11 g 2):

  call the query of disc elapsed to cpu count current lines

  ------- ------  -------- ---------- ---------- ---------- ----------  ----------

  Parse        1      0.00       0.00          0          0          0           0

  Run 1 0.04 0.02 0 87 0 0

  Get 9 1.96 7.81 65957 174756 0 873

  ------- ------  -------- ---------- ---------- ---------- ----------  ----------

  total 11 2.01 7.84 65957 174843 0 873

  Elapsed time are waiting on the following events:

  Event waited on times max wait for the Total WHEREAS

  ----------------------------------------   Waited  ----------  ------------

  SQL * Net message to client 9 0.00 0.00

  reliable message 1 0.00 0.00

  ENQ: KO - checkpoint fast object 1 0.00 0.00

  direct path read 5074 0.05 5,88

  SQL * Net more data to the customer 5 0.00 0.00

  SQL * Net client message 9 0.01 0.00

  ********************************************************************************

  We can see that the 65957 physical disk reads resulted only 5074 direct path read. Normally, the number of physical reads from disk is more directly reflected in the wait events section.

  Is this normal? Why is that? Maybe because these discs are on a San have a cache?

  Best regards.

  Carl

  direct path read is an operation of e/s diluvium, is not just to read 1 block

• question about physical reads

  RMAN I/o is taken into account in the stat of physical reads ?
  
  I mean, it's blocks read from disk by RMAN to backup the database is represented in physical reads statistics?
  
  One of our databases has a value for the IO requests total physical reading greater than physical reads , so I guess that the missing physical reads are RMAN.
  
  Thanks in advance

  blocks read from disk by RMAN to back up the database represented by physical reads statistics?

  Yes, I do. It is something that can be tested by looking at the session for the RMAN sessions statistics.

  Hemant K Collette

• How to check the instance datafile I/O? solve the high physical reads/writes?

  Method-1
  identify 'hot spots' or I/O contention
  Select NAME,
  "Physical reads," PHYRDS
  round ((PHYRDS/PD.)) (PHYS_READS) * 100, 2) «Read %»,
  PHYWRTS "physical Scriptures."
  Round (PHYWRTS * 100/PD.) PHYS_WRTS, 2) "% of writing."
  FS. PHYBLKRD + FS. PHYBLKWRT "Total block I / O of.
  de)
  Select sum (PHYRDS) PHYS_READS,
  Sum (PHYWRTS) PHYS_WRTS
  v $ filestat
  ) pd,.
  v$ datafile df,.
  v$ filestat fs
  where df. FILE # = fs. FILE NO.
  ranking by fs. PHYBLKRD + fs. PHYBLKWRT / / desc
  
  
  Another method-
  On Oracle10g, CWA also provides the dba_hist_filestatxs table to track the disk i/o:
  
  break on begin_interval_time jump 2
  
  column phyrds 999 999 999 format
  column begin_interval_time format a25
  
  Select
  begin_interval_time,
  file name,
  phyrds
  Of
  dba_hist_filestatxs
  natural join
  dba_hist_snapshot;
  
  It is - method that you use to check for errors in data I/O file? AND how to solve high physical reads and writes?

  Ankit Ashok Aggarwal wrote:
  AWR project stat e/s in terms of segments, query and etc.
  Here, my concern is what should be the value of ideal threashold for a DBA to act on the high physical reads/writes?
  and how he can fix this problem if it affects database performance?

  I don't think that there is any possibility to mention that the amount of an e/s is going to be good or bad for a database in general. The reason is that it is not quantifiable. For example, in a store, maybe one day that no one gets and another day, there is no place to stand (because there was a sale announced by the store). So now, how do you say that it was not necessary? Rather than search for best practice values, it is best that you keep track of the normal functioning of your db and when you see a special summit in the values of events scattered user IO as DB file read, DB file parallel read etc and when this happens, you should check back the work you are doing and see if that can be refined somehow.

  Aman...

• more physical reads the blocks (in this partition)

  Hi all
  
  I have the partitioned table hash on an indexed column 'id', which is non-unique and a part of my primary key. Inside each partition, lines with the same id are located close to each other which is done by dbms_redefinition reorg using orderby_cols. The intention is to reduce the amount of physical reads as there are no requests that don't filter on the column id.
  
  However, what I see is a large number of physical reads. The first partition has roughly 80K lines, a length of 347, block size of 8K and compression mean line... resulting in 821 blocks. And when (after flushing cache in the memory buffer and shared pool) submit a query that filters on 'id' only and results is 106 selected lines, I see to roughly 1400 physical reads.
  
  --------------------------------------------------------------------------------------------------------------------------------
  | ID | Operation | Name | Lines | Bytes | Cost (% CPU). Time | Pstart. Pstop |
  --------------------------------------------------------------------------------------------------------------------------------
  | 0 | SELECT STATEMENT | 106. 36782 | 3 (0) | 00:00:01 |
  | 1. SIMPLE HASH PARTITION | 106. 36782 | 3 (0) | 00:00:01 | 1. 1.
  | 2. TABLE ACCESS BY LOCAL INDEX ROWID | XXX MOVIES | 106. 36782 | 3 (0) | 00:00:01 | 1. 1.
  |* 3 | INDEX RANGE SCAN | XXXXX | 1 | | 1 (0) | 00:00:01 | 1. 1.
  --------------------------------------------------------------------------------------------------------------------------------
  
  Information of predicates (identified by the operation identity card):
  ---------------------------------------------------
  
  3 - access ("ID" = 49743)
  
  
  Statistics
  ----------------------------------------------------------
  22243 recursive calls
  0 db block Gets
  66651 compatible Gets
  physical reads 1404
  0 redo size
  10933 bytes sent via SQL * Net to client
  299 bytes received via SQL * Net from client
  9 SQL * Net back and forth to and from the client
  150 sorts (memory)
  0 sorts (disk)
  106 lines processed
  
  
  I expect to see around 10-15 physical reads. And I don't understand how it can be 1400 physical reads with only 821 blocks in this partition. Can someone explain this behavior to me?
  
  Thanks in advance,
  Dirk

  DRabbit wrote:
  However, what I see is a large number of physical reads. The first partition has roughly 80K lines, a length of 347, block size of 8K and compression mean line... resulting in 821 blocks. And when (after flushing cache in the memory buffer and shared pool) submit a query that filters on 'id' only and results is 106 selected lines, I see to roughly 1400 physical reads.

  --------------------------------------------------------------------------------------------------------------------------------
  | Id  | Operation                          | Name  | Rows  | Bytes | Cost (%CPU)| Time     | Pstart| Pstop |
  --------------------------------------------------------------------------------------------------------------------------------
  |   0 | SELECT STATEMENT                   |       |   106 | 36782 |     3   (0)| 00:00:01 |       |       |
  |   1 |  PARTITION HASH SINGLE             |       |   106 | 36782 |     3   (0)| 00:00:01 |     1 |     1 |
  |   2 |   TABLE ACCESS BY LOCAL INDEX ROWID| XXX   |   106 | 36782 |     3   (0)| 00:00:01 |     1 |     1 |
  |*  3 |    INDEX RANGE SCAN                | XXXXX |     1 |       |     1   (0)| 00:00:01 |     1 |     1 |
  --------------------------------------------------------------------------------------------------------------------------------
  
  Predicate Information (identified by operation id):
  ---------------------------------------------------
  
  3 - access("ID"=49743)
  
  Statistiken
  ----------------------------------------------------------
  22243  recursive calls
  0  db block gets
  66651  consistent gets
  1404  physical reads
  0  redo size
  10933  bytes sent via SQL*Net to client
  299  bytes received via SQL*Net from client
  9  SQL*Net roundtrips to/from client
  150  sorts (memory)
  0  sorts (disk)
  106  rows processed
  

  For me, the most important statistic is recursive calls. Even for a partitioned table with (for example) a few thousands of partitions, this seems very high for a simple analysis and perform call that eventually access a single partition known by index.

  Test strategies - rerun the query immediately to see if you get similar results, run the query again ONLY after a buffer_cache (not shared_pool) hunt.
  It is possible that more work on optimization, or calls to (for example) a pl/sql function embedded. The behavior of the repeated executions can give you a clue on this subject.
  (And if none of these options can help, then activation trace and check the trace file should help).

  Concerning
  Jonathan Lewis

• Physical read

  Hello
  
  Read how to calculate physical and logical to read?
  
  THX.

  Hello

  You can have a quick glance at the V$ SEGMENT_STATISTICS and the filter on the STATISTIC_NAME column.

  For example, if you want physical reads and logical reads about a TABLETEST segment (for example :)), you can use this query:

  select OBJECT_NAME,SUBOBJECT_NAME,statistic_name,value
  from V$SEGMENT_STATISTICS
  where statistic_name in ('physical reads','logical reads')
  order by 5
  

  If you want historical data, you can take a look at the DBA_HIST_SEG_STAT view.

  Laurent

• Physical read of Checkpoint/T-Logs of TimesTen IMDB

  My data store has forced (vivid RAM) memory, and so I want to flush the data from IMDB to his Checkpoint/T-logs file with the idea to free the memory.
  
  Yes, is there a any way by which data in IMDB can be deleted (freeing the memory), and on request, the data can be loaded into IMDB? Something like a physical disc of Oracle reading. I am aware that it is possible with the IMCD with aging concept. But my setup has no underlying oracle.
  
  In other words, I want data from IMDB to be obsolete, and I need to persistent storage (checkpoint/T-logs) on request. First part of the question (age-out) in IMDB seems to be possible... and I need confirmation on the second part of the question.
  
  Create new data for constraint violations store memory is not an option that the constraint of the memory is at the host level and not at the data store level.
  
  Published by: Jeganlal on December 9, 2010 11:14

  No it's not possible using standalone TimesTen. TimesTen is a database in memory, and there is no concept of some data only resides on the disk. The checkpoint field and newspapers are there purely for the dat in memory of persistence and transactionality. Either you need more memory or you need to switch to a scenario of IMDB Cache rather than standalone TImesTen.

  Chris

• Top 10 SQL by physical reads

  Dear experts,
  
  Version of Oracle - 11.1.0.7
  
  Do we not have a way to draw 10 albums SQL (physical readings) for a period of 24 hours based on the AWR data or other points of view V$.
  
  Thank you

  DBA_HIST_SQLSTAT, all you can say is that was the user of the analysis.
  Not the user running.

  DBA_HIST_SQLSTAT is therefore insufficient for queries that are executed several times by several users, and there is not a lot of solution.

  You could have a glance in the underlying data from the ASHES - V$ ACTIVE_SESSION_HISTORY and DBA_HIST_ACTIVE_SESS_HISTORY which is respectively sampled data and a sample of a sample and it could give an indication of who is running.

  Published by: Dom Brooks on July 19, 2012 10:18
  Part of sentence was missing

• Physical reads in the data base

  Dear Experts,
  
  I need to get separate SQL ID for the SQLs executed by a particular user - "PROD" b/w 09:00-17:00 (snap ID 1000-1008). I wrote below SQL:
  
  Select sql_id distinct from DBA_HIST_SQLSTAT where PARSING_SCHEMA_NAME = "PROD" and snap_id between 1000 and 1008
  
  Everyone sees a reason why this can be bad?
  
  Thanks for sharing of the entries.

  Select sql_id distinct from DBA_HIST_SQLSTAT where PARSING_SCHEMA_NAME = "PROD" and snap_id between 1000 and 1008
  Everyone sees a reason why this can be bad?

  request looks good, it makes you want you need.

• read tip or parameter to force physics

  I'm using Oracle 11.2.0.3. I have a query that took 45 minutes the first time and it take 4 minutes to subsequent term in QC environment. In both cases he uses the same plan. If I try the query again in a few days, first time, it takes a lot of time. Most of the waiting is in scan limited index range - "parallel reading of file db.

  Same query is executed within 2 minutes in a down environment with different plan. I used the tip to make the plan even in the environment of QC. Now query executes as expected, but I suspect he might slow down if the data is not cached. I do not have access to change the privileges of system buffer cache flush.

  Is there a trick or the setting I can use physical force read?

  What view will tell me if a table is always cached in memory?

  I'm not sure that there is no parameter to force the physical reading. In addition to the view V$ BH, you can take a look at the output of the command SET AUTOTRACE as well. It would show if there are physics involved in the execution of the query or not bed. But for blocks, the view V$ BH would be useful. I would also like to add that a request would certainly be difference in performance after 1st run. That's why the "warm-up of the buffer cache" is recommended to be done by running the query at least 2 times before you start measuring its performance.

  Aman...

• Difference between ' Please read physical Total i/o per second "and"physical IO read queries per second.

  What is the difference between "Physics read Total IO requests by s ' and ' physical IO read requests per s ' metric in dba_hist_sysmetric_summary?

  Documentation (reference, Descriptions of statistics):

  IO physical read requests : number of read requests for application activity (mainly the buffer cache and direct load operation) that read one or more blocks of data by request. It is a subset of statistics 'physical read IO total number of applications.

  requests for IO total physical reading : number of read requests that read one or more blocks of data base for any activity, including the application instance, backup and restore, and other public services.

• Impossible to make the link between the consistent readings with the number of physical block reads

  Hello
  
  The issue is that we have seen consistent readings are more than the total buffers necessary to give back the results.
  I emptied the buffer_cache before you execute the query and also questioned the V$ BH to know the details of the buffer for these objects... After the flush before you trigger the request that we do not have buffers about these tables. Which is expected.
  We are doing sequential reads of DB file with the plan and will result in a single block, read at a time.
  Please take a close look at "TABLE ACCESS BY INDEX ROWID CMPGN_DIM (cr = pr 45379 = 22949 pw = time 0 = 52434931 US)" line in the plan source line below...
  Here we have only 22949 data buffers 22949 means reading physical but we are witnessing gets 45379 coherent.
  
  Note: We have the CMPGN_DIM and AD_GRP tables in the tablespace size 4 M block and we have only the default db_cache_size. My database block size is 8192.
  
  Can you please help me understand how the 22949 sequential reads the result in 45379 coherent takes.
  
  Even the V$ BH query buffer details matches with physical reads.
  
  
  line source of 10043 trace query plan:
  
  SORT ORDER BY 27 (cr = 92355 pr 47396 pw = time = 0 = 359030364 en)
  27 WINDOW SORT (cr = 92355 pr 47396 pw = time = 0 = 359030088 en)
  NESTED LOOPS EXTRA-ATMOSPHERIQUE 27 (cr = 92355 pr 47396 pw = time = 0 = 359094569 en)
  NESTED LOOPS EXTRA-ATMOSPHERIQUE 27 (cr = 92276 pr 47395 pw = time = 0 = 359041825 en)
  VIEW 27 (cr = 92197 pr 47393 pw = time = 0 = 358984314 en)
  UNION-ALL 27 (cr = 92197 pr 47393 pw = time = 0 = 358984120 en)
  HASH GROUP BY 26 (cr = 92197 pr 47393 pw = time = 0 = 358983665 en)
  VIEW 9400 (cr = 92197 pr 47393 pw = time = 0 = 359094286 en)
  9400 COUNT (cr = 92197 pr 47393 pw = time = 0 = 359056676 en)
  VIEW 9400 (cr = 92197 pr 47393 pw = time = 0 = 359009672 en)
  9400 SORT ORDER BY (cr = 92197 pr 47393 pw = time = 0 = 358972063 en)
  9400 HASH JOIN EXTERNAL (cr = 92197 pr 47393 pw = time = 0 = 358954170 en)
  VIEW 9400 (cr = 92191 pr 47387 pw = time = 0 = 349796124 en)
  JOIN by HASH 9400 (cr = 92191 pr 47387 pw = time = 0 = 349758517 en)
  94 the TABLE ACCESS BY INDEX ROWID CMPGN_DIM (cr = pr 45379 = 22949 pw = time 0 = 52434931 US)
  50700 IDX_CMPGN_DIM_UK1 INDEX RANGE SCAN (cr = 351 pr = 349 pw = time 0 = 1915239 US)(object id 55617)
  TABLE ACCESS BY INDEX ROWID AD_GRP 60335 (cr = 46812 pr 24438 pw = time = 0 = 208234661 en)
  60335 INDEX RANGE SCAN IDX_AD_GRP2 (cr = 613 pr = 611 pw = time 0 = 13350221 US)(object id 10072801)
  VIEW 7 (cr = 6 pr = 6 pw = time 0 = 72933 en)
  7 HASH GROUP BY (cr = 6 pr = pw 6 = 0, time = 72898 US)
  162 PARTITION SINGLE PARTITION RANGE: 4 4 (cr = 6 pr = 6 pw = time 0 = 45363 en)
  162 PARTITION PARTITION UNIQUE HASH: 676 676 (cr = 6 pr = pw 6 = 0, time = 44690 US)
  162 INDEX RANGE SCAN PK_AD_GRP_DTL_FACT PARTITION: 3748 3748 (cr = 6 pr = pw 6 = 0, time = 44031 US) (object id 8347241)
  1 DOUBLE QUICK (cr = 0 pr = 0 pw = time 0 = US 9)
  25 TABLE ACCESS BY INDEX ROWID AD_GRP (cr = 79 pr = 2 pw = time 0 = 29817 US)

  user622572 wrote:
  Hi Charles,
  In this test of problem we have not any activity on my tables other than my session. If we do not have all cancel must be applied to the brand CR.

  * "So, which means the above?" *-->, that means the number of times where the blocks was visited is counted as CR not the number of buffers required to respond to a request. "

  * "" Why a plan of execution would change the number of makes more sense? "*--> changes in the method of access to the path and join required a different number of times the same buffer necessary visit." It means NL require more number of same buffer visits to T2, T1 and INT_T1_C2 as the hash join.

  * "' Why not flush the buffer cache would result in the PR drop statistics to 0, but does not change the statistics CR?" *--> each result buffer in physics RAS reads for objects from there which is why it cannot be zero in the first 4 tests where as test5 require PI (we don't rinse prior to that in your test case). " CR cannot be changed with buffer flush. Required CR same amount of pads and tampons visits regardless of buffer flush.

  It seems that we can say that "CR numbers include the number of times the block was visited in a query instead of just the number of buffers required". ""

  user622572,

  You deserve a lot of credit for having read and understood what I posted to correctly answer the questions I asked at the bottom of results and for your ability to answer your original question.

  Charles Hooper
  IT Manager/Oracle DBA
  K & M-making Machine, Inc.

• Details on the logical read in oracle

  The difference between logical and physical read in oracle is that in the case of physical read data block is read from the data located on the disk files as in the case of logical read data block is read from buffer cache already present it... is the concept of right or need more refinement...

  You are right. A physical read from disk, a logical reading is in the cache.

  I would also like to add that some people look at wrong only physical reads. I had to set a lot of queries that have carried out a number of face of logical reads. Logical reads are still work. A query that is running 1 million logical reads to return 100 lines is probably a request requiring adjustment.

  See you soon,.
  Brian

• An insert in TWG read a lot of UNDO

  Hello

  I have an insert in a TWG (whose one index for its primary key) that is read a lot of UNDO (db file sequential read file #= 2).

  I can't think of a good reason for it.

  Here's a tkprof:

  SQL ID: hash of Plan 4xwgpzxtgg4s5: 0

  INSERT INTO xxxxx.xxxxxxxxxxxxxxxxxx (INCOUNT, INNUM)

  VALUES

  (: v1,: v2)

  call the query of disc elapsed to cpu count current lines

  ------- ------  -------- ---------- ---------- ---------- ----------  ----------

  Parse        0      0.00       0.00          0          0          0           0

  Run the 113 0.09 0.09 36 173 838 6830

  Fetch        0      0.00       0.00          0          0          0           0

  ------- ------  -------- ---------- ---------- ---------- ----------  ----------

  Total 113 0.09 0.09 36 173 838 6830

  Chess in the library during parsing cache: 0

  Optimizer mode: ALL_ROWS

  The analysis of the user id: 108

  Elapsed time are waiting on the following events:

  Event waited on times max wait for the Total WHEREAS

  ----------------------------------------   Waited  ----------  ------------

  SQL * Net message to client 17 0.00 0.00

  SQL * Net client message 17 0.01 0.04

  DB file sequential read 36 0.05 0.02

  All are file #= 2:

  WAITING #2390071304: nam = 'db file sequential read' ela = 735 file #= 2 block # 46834 blocks = 1 obj #= 0 tim = 3960230031525

  WAITING #2390071304: nam = 'db file sequential read' ela = 713 file #= 2 #= 46835 blocks = 1 obj #= 0 tim = 3960230032327

  WAITING #2390071304: nam = 'db file sequential read' ela = 659 file #= 2 #= 46836 blocks = 1 obj #= 0 tim = 3960230033125

  WAITING #2390071304: nam = 'db file sequential read' ela = 648 file #= 2 block # 46837 blocks = 1 obj #= 0 tim = 3960230034030

  EXEC #2390071304:c = 0, e = 3535, p = 4, cr = 5, cu = 38, SIG = 0, r is 52, dep = 0, og = 1, plh = 0, tim = 3960230034104

  And there is another thing I don't understand. Each EXEC has only a single LINK, such as:

  LINKS FOR #2390071304:

  Link #0

  oacdty = 02 mxl = 22 (22) mxlc = 00 wrong = 00 = 00 = 00 pre scl

  oacflg = 01 fl2 = 1000000 frm = 00 csi = 00 siz = 48 off = 0

  kxsbbbfp = bln 8e785608 = 22 = 02 = 05 flg avl

  value = 12

  Link #1

  oacdty = 02 mxl = 22 (22) mxlc = 00 wrong = 00 = 00 = 00 pre scl

  oacflg = 01 fl2 = 1000000 frm = 00 csi = 00 siz = off 0 = 24

  kxsbbbfp = bln 8e785620 = 22 avl = flg 05 = 01

  value = 7961958

  However, the execs reported about 50 lines (as you can see the number of lines in the tkprof)

  Any idea of the reason for this?

  Thank you very much

  Kind regards

  Franck.

  Check if the recording of flashback is enabled.

  Others explain the curiosity to see a single set of values of bind variables when you table inserts - playback of cancellation could be "physical reads for new return of flame." I have provided a brief description in the last comment to the next note of the RTO: Flashback - high load

  Concerning

  Jonathan Lewis