OBIEE 11 g performance tuning

Similar Questions

• Foglight performance tuning - Cache Code JVM

  Just curious about this setting. Currently, our new FMS are amazing. But I noticed this option in the field guide. No indication on that? The docs a bit thin. When I use it? What is the default if I don't specify? How to monitor the need to increase this?

  Performance Tuning field Guide

  To optimize the execution of code in larger environments, you can increase the size of the

  JVM code cache.

  To increase the size of the JVM code cache:

  1 open /config/foglight.config.

  2. change the line foglight.vm.option0 as follows:

  Foglight.VM.option0 = '-XX:ReservedCodeCacheSize = 128 m "

  3. restart the management server.

  The default setting is different based on the platform of the JVM.

  It is default to 1024 m on 64-bit systems or 32 m on most systems except Solaris where it could be 48 m.

  It sets the size used by the cache code during compilation JIT compiler. If this cache is small you will receive errors like this in the log file

  Java HotSpot(TM) Client VM warning: CodeCache is full. Compiler has been disabled. Java HotSpot(TM) Client VM warning: Try increasing the code cache size using -XXReservedCodeCacheSize=

  And it allows to disable the JIT compiler, which will result in reduced performance because all dynamic logic as rules and other groovy scripts will no longer be dealt with by the JIT.

  The default settings will work in most cases, but a big FMS with many objects and rules may require a greater reserve of cache to keep the happy JIT compiler.

  Stefan

• can I use "Oracle Database 12 c: performance management and Tuning" training for the certification "Oracle Database 11g: Performance Tuning 1Z0-054 '"»

  I took "Oracle Database 12 c: new performance management and Tuning" the oracle University training. Now I would like to get certified on "Oracle Database 11g: Performance Tuning 1Z0-054 ' exam. Is this possible?

  I guess you ask if you can use the course 12 c as long as the condition of course for the review of 11g.  Over 12 c is not listed as one of the options for the 11 g certification and course requirements are normally specific version - at least with DBA certifications.  If you are already an Oracle OCP DBA, of course, there is no requirement of course for the review of performance tuning.  From what I know the training requirements for other certifications, I do not that you will be able to use it. However, Brandye will provide a definitive answer to whether the course 12 c would be acceptable for 11g certification.

  That said, I'm with John - 12 c review is about a community of 85 to 90% in the review of the 11g and is currently about 20% of the price while it is in beta.  What is the point of trying to cross the releases?

• 1Z0-054: Oracle Database 11g Performance Tuning Expert

  I intend to give the exam 1z0-054 for the certification, "Oracle Database 11g Performance Tuning Expert".
  
  The path to Certification indicates that, if you are 'OCP DBA 11 g' then you can directly take this article and there is no requirement for training instructor. Therefore, I decided to prepare my own. Can someone guide me on study material and landfills?
  
  Kind regards
  
  Ahmer Mansoor

  Therefore, I decided to prepare my own. Can someone guide me on study material and landfills?

  First of all, the word "shock" has a precise meaning. It is short for "landfills of brain", which are illegal and can cost you this certification and all others you have taken.

  http://blogs.Oracle.com/certification/entry/0477

  Second - there is currently no study guide or a specific book to this certification at the moment. I write the study guides and this one is actually next on my list when I finished the current book to * 1Z0-117: Oracle Database 11 g Release 2: SQL Tuning *. I have a guide for 1Z0 - 054 before sometime in may more likely. That said, I have a few links for legal study on the companion Web site material for my series of guide to the subjects for this exam. It also has links to the Oracle documentation that contains the information you need. Once I have started writing the book review, links will develop quickly because I research for her themes and links to the documentation could extend as well.

  http://www.oraclecertificationprep.com/Apex/f?p=OCPSG:EXAM_DETAILS:0:no:P2_EXAM:1Z0 - 054

• Will there be any tools avialable to the on query performance tuning

  Hello
  
  How to tune a database query?
  Suppose that there are basic query "select * from emp where empid = '1101' ';"
  
  Y at - it all avilabale tools to suggest performance Tuning on a query?
  
  How can I resolve this application, and what is the best approach on this?
  
  Please share your ideas and thanks.

  Oracle provides tools such as SQL Trace and TKPROF. For the information on a SQL query.

  But basically, you should know how oracle works. You must be aware of the different access plans and know why they are being used.

  You need to know when to use an index and when not to use a.

  There are many more. Best thing for you would be to start reading the document.

• Oracle Application Performance Tuning

  Hi all
  Please forgive me if I'm asking this in the wrong section.
  
  I'm an engineer s/w with 2 years exp. I worked as a tuning performance for Oracle EBS resource in my company. Jobs in question especially SQL tuning and treating indices etc. At the moment I took interest in s/n things, and I finished my TRAY in Oracle 10 g. Now, my company gives me the opportunity to move to a team of DBA of the Application and I'm totally confused about this.
  
  Becoming a DBA Apps means that the effort that I put in the certification in 10g will be of no use. There are other Apps DBA certification documents. I should stay put in performance tuning & wait for a chance to base DBA or I will accept the opportunity to Apps DBA.
  
  Also, did my exp. in SQL tuning performance hold any value as such with exposure to activities of s/n?
  
  Please guide me in this matter I am very confused right now.
  
  Kind regards
  Lucette

  Lucette wrote:
  Hello bigdelboy, thank you very much for your answer.

  Yes, I meant by oracle Apps DBA Oracle EBS.

  1Z0-046 clearing, is an option. However, my question is will delete both reviews the Admin I and Admin II being of any help, bypassing the expericnce practice? The EBS DBA work involves support and patching, cloning and installation of the new node etc.

  Also, is my performance tuning experience will help me somehow on the way forward as a DBA / EBS DBA?

  Thank you for your valuable time.

  Kind regards
  Lucette Sarath

  The way I read it is that.

  People notice that you are able to understand and SQL performance. And you must have some knowledge of Oracle EBS. And in fact, you also have a DBA OCA.

  So, there are 98% + chance you can be transformed into a good Oracle Apps DBA (and the core s/n). If I was in their position I want you on my team. It's the way we used to bring on DBA in the old days before approval, and he still has a lot of merit. OK you can only do tasks limited at first... but the number of taks you can do will increase substantially in recent months.

  I imagine that the Oracle Apps DBA will be pleased to have a person on board who can tune SQL which sometimes can sometimes seem more like an art than a science of performance. The application of patches, etc. can be taught in small doses. If they are a good team, they're not trying to give you all at once, they'll get to learn one procedure at a time.

  And remember, if in doubt ask. and if you do not understand ask again. And be safe and not sorry on actions that could be serious.

  .................

  If your fear on liinux: recipes of Linux for Oracle DBA (Apress) might be a good book to read but can be expensive in India
  Same: (Oracle Applications Dba field Guide) can agree, and the same (Rman recipes for Oracle Database 11 g A problem-solution approach) can have some value if you need to get fast speed in these areas.
  They are not perfect, but they sometimes have to consolidate the information carefully; However only buy if you think they are pretty cheap. You can buy them well and feel disappointed (these all is found by Apress which have produced a number of good books... they also published some waste as well)

  And review the dba examples 2 days as well and install examples n OTN linux so... they are free from books I mentioned.

  Rgds - bigdelboy.

• Dates for "Oracle Database 11g: Performance Tuning" exam

  Hello!
  
  Today Oracle posted the info on the new database Oracle 11 g Performance Tuning review required for new certification expertise:
  
  http://education.Oracle.com/pls/web_prod-PLQ-dad/db_pages.GetPage?page_id=41 & p_exam_id = 1Z0_054 #2
  
  It is said that pre-registration for the beta version review starts today, however, it does not say when the phase beta begins and ends. Anyone know?
  
  Thank you
  Marcus

  March 31, 2009 is the current provisional end date for the exam in beta 1Z1-054. This date is subject to change.

  Kind regards
  Brandye Barrington
  Certification Forum Moderator
  Manager certification program

• 1Z0-024 1Z0-054 vs Performance Tuning

  Hello

  obiously, they are tests for very different versions (8i vs 11g), but if you take that into consideration, I think they are very similar... But in Oracle 8i you had to pass 5 exams (Yes FIVE!) If you want to get your OCP. Today, you get 11G OCP with 3 tests... and if you pass 1Z0-054 you have one cerfitication others...

  Is this true?

  Are different so the two reviews?, it seems that they are very similar (all similar which can be oracle 8i and oracle 11g)

  Anyone who has had 1z0-024 (there are almost 13 years) took 1z0-054?

  Concerning

  Is this true?

  I guess that what you are asking to be true is that the 11g OCP requires three exams (TRUE) and passing the 1Z0 - 054 you grant OCÉ certification in addition to the 11 g OCP (also TRUE).  Theoretically, you could also win the 11g tuning certification without an OCP, but there're so much, we need practical training that must also be met.

  Are different so the two reviews?, it seems that they are very similar (all similar which can be oracle 8i and oracle 11g)

  Well, I don't have a handy list of topics for 1Z0 - 024, so I'll work from subjects for 1Z0 - 054 and extrapolate. First of all, I will list some of the topics in 054 which are features that did not exist in 8i:

  Automatic maintenance AWR, ADDM, tasks, reports of ASH, database Resource manager, planner of the Oracle, SQL tuning automatically, planning Plan SQL, SQL Performance Analyzer, Database Replay, Automatic Shared Memory Management, automatic memory management, space management auto Segment, Segment retractable, Automatic Storage Management, Statspack.

  Please note that the list above is perhaps not complete.  I might have missed some.  However, I think I can answer your question with confidence, saying: "Yes - the two reviews are very different."

• How to use the advice of Obiee to increase performance

  Someone please tell me how can we use tips to increase the performance of ad-hoc and obiee dashboard reports

  Hello

  Check it,

  http://www.howtoexam.com/index.php?option=com_content&view=article&ID=75%3Ausing-hints-in-OBIEE-RPD-and-answers&CATID=790%3Acomputers-and-software&Itemid=166

  Rgds,
  DpKa

• Experience Performance Tuning Oracle professional

  Hi people,
  
  has followed the performance oracle tuning professional experience can be used to replace the certificate of apprenticeship or training in the OCP 9i - DBA?
  
  concerning
  Sunil

  sunilkrishna wrote:
  Hi people,

  has followed the performance oracle tuning professional experience can be used to replace the certificate of apprenticeship or training in the OCP 9i - DBA?

  concerning
  Sunil

  No, if you want proof of the OCP DBA, you will need classroom training. Tip: WDPFund, it may be cheaper. Or do you have a 11g s/n of class performance... you can learn something new.

• Difficulty of fix DBA Performance Tuning applications

  Hello
  I had an interview, which the DBA team leader asked me about my expereince of performance
  
  I met on my expereince in the sqls of tuning and get a good statistic so the CBO choose the best plan, or even hint my sqls so I force the best plan
  
  the interviwer was worried because he is an application development, does not a DBA tuning
  mention of He that he wanted to hear that I've changed a setting db that improve performance, or change something in the DBA team that optimize performance
  
  Does anyone have an example of changing the setting or something DBAish that helps the performance?
  
  Thanks for any input

  Index rebuild or change the block size?

  (game of lights and stands back)

  Choose a parameter, a parameter, for example:
  How about increasing size of log file to reduce log file switches?
  With the help of async_io?
  Change of db_multiblock_io_count, optimizer_index_adj.
  Going between manual and automatic memory management.
  The fast setting = true ;) etc, etc.

• Performance tuning for Unix hosted components

  Hello

  I have Foundation, Planning, APS, EAS on Unix - Solaris environment. Do I need to increase the size of the JAVA heap for these components? We do this on windows, but I don't remember done previously for OEL.  It is meaningless for me to increase java heap xms and xmx values of size of 128 to 4096.

  version of hyp: 11.1.2.4

  Please advice

  Thank you

  Anjum

  Allow you the JVM, whether windows or * nix platforms, although saying that according to the load it may not need passing in 11.1.2.4 as the xmx will be 4096

  It is interesting to go through the tuning guide- https://blogs.oracle.com/pa/resource/Oracle_EPM_11_1_2_3_Tuning_Guide_v2.pdf

  Even if it's for 11.1.2.3 it is still topical for most in 11.1.2.4

  See you soon

  John

• FDM: Import performance Tuning, where "import processes: Map" is more times that takes the report

  Hi Experts,

  FDM report shows the "import process: map ' activity is much more time during the import process completed by do from the web interface for a unique time period (file 1 month = 4 MB). Please suggest how I can improve it?

  Thank you

  Vivek

  In this doc attached are details on parameters DB contribute to improve performance.

  http://docs.Oracle.com/CD/E40248_01/EPM.1112/fdm_dba.PDF

  A tip, I would also give is the importance of the order of mapping. Basically, if you like the mappings you should order based on volume. If a similar mapping affects a large volume of documents, you want to apply only the first. This allows to reduce the Recordset down over time so that the least used mappings go against the smallest possible demappe data set.

• Help with PL/SQL Performance Tuning

  Hi all

  I have a PL/SQL procedure, it works very well. No error and no bugs. However its still holding at the end I use the concatenation operator (|), and I know that its expensive. How to improve the performance of the procedure?

  Here is the code

  create or replace 
  PROCEDURE POST_ADDRESS_CLEANSE AS
  
  
  CURSOR C1 IS
  SELECT Z.ROW_ID,
          Z.NAME
  FROM  STGDATA.ACCOUNT_SOURCE Z;
  
  
  CURSOR  C2 IS 
  SELECT  DISTINCT CLEANSED_NAME || CLEANSED_STREET_ADDRESS || 
          CLEANSED_STREET_ADDRESS_2 || CLEANSED_CITY || CLEANSED_STATE || 
          CLEANSED_POSTAL_CODE AS FULLRECORD
  FROM    STGDATA.ACCOUNT_SOURCE_CLEANSED;
  
  
  V_ROWID Number := 1;
  V_FLAG VARCHAR2(30);
  TEMP_ROW_ID VARCHAR2(10) := NULL;
  
  
  BEGIN
  
  
    -- This loop will update CLEANSED_NAME column in ACCOUNT_SOURCE_CLEANSED table.
    FOR X IN C1 LOOP
      
      TEMP_ROW_ID := TO_CHAR(X.ROW_ID);
      
    UPDATE STGDATA.ACCOUNT_SOURCE_CLEANSED A
    SET  A.CLEANSED_NAME = X.NAME
    WHERE A.ROW_ID = TEMP_ROW_ID;
      
      COMMIT;
  
    END LOOP;
    
    -- This loop will update columns EM_PRIMARY_FLAG, EM_GROUP_ID in ACCOUNT_SOURCE_CLEANSED table
    FOR Y IN C2 LOOP
  
  
      UPDATE  STGDATA.ACCOUNT_SOURCE_CLEANSED
      SET     EM_GROUP_ID = V_ROWID
      WHERE   CLEANSED_NAME || CLEANSED_STREET_ADDRESS || CLEANSED_STREET_ADDRESS_2 || 
              CLEANSED_CITY || CLEANSED_STATE || CLEANSED_POSTAL_CODE = Y.FULLRECORD;
  
  
      UPDATE  STGDATA.ACCOUNT_SOURCE_CLEANSED
      SET     EM_PRIMARY_FLAG = 'Y'
      WHERE   CLEANSED_NAME || CLEANSED_STREET_ADDRESS || CLEANSED_STREET_ADDRESS_2 || 
              CLEANSED_CITY || CLEANSED_STATE || CLEANSED_POSTAL_CODE = Y.FULLRECORD 
      AND     ROWNUM = 1;
      
      V_ROWID := V_ROWID + 1;
  
  
      COMMIT;
      
    END LOOP;
    
    UPDATE  STGDATA.ACCOUNT_SOURCE_CLEANSED
    SET     EM_PRIMARY_FLAG = 'N'
    WHERE   EM_PRIMARY_FLAG IS NULL;
    
    COMMIT;
  
  
    --dbms_output.put_line('V_ROW:'||V_ROWID);
    --dbms_output.put_line('CLEANSED_NAME:'||Y.FULLRECORD);  
    
  END POST_ADDRESS_CLEANSE;
  

  Thanks in advance.

  Post edited by: Rooney - added code using the syntax highlight

  Thanks for everyone entered.

  I was able to solve the problem. My solution using the old code with the |, I was able to create an index on the following attributes:

  CLEANSED_NAME | CLEANSED_STREET_ADDRESS |  CLEANSED_STREET_ADDRESS_2 | CLEANSED_CITY | CLEANSED_STATE | CLEANSED_POSTAL_CODE

  I never knew that you can create an index on the concatenated attributes sets. Doing this, I was able to update all of my files and improve performance. All records ran in 80 seconds.

  Thanks again for the help.

• Application performance tuning

  Hello

  
  The Oracle version:-10.2.0.5

  I post the question in SQL and PL/SQL Forum, but have received no answer, so posting here.

  
  We tried to address a performance problem in our system, which is a layer of ETL to another application and receives data in the form of flat files on a daily basis. The received data are loaded into a temporary, not physically temporary table but appointed to keep the data for 2 days, after which it will be deleted. Late in the day, all the data is copied from temporary tables to the actual tables, which have a 5 days retention policy.

  
  Many medium-sized lines worked for each day is 1.8 to 2 million. And the number of lines for the most part remains stable in each of the two tables.

  There are 5 packages, each containing about 3 applications similar to those mentioned below (with minor differences) and there are about 5-7 jobs to execute the package at a time. This amounts to 15-20 performance of both statements.

  The AWR report

  Name of the statistic	Time (s)	% of time of DB
  Execute SQL elapsed time	15,807.62	97.12
  DB CPU	9,667.99	room 59,40

  This is the number of data in the two tables is as below:

  OUT_TABLE OUT_TEMP

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

  8763750 5255109

  Each of these tables is Composite partition-ed, primary partition Date (range) and the partition of Sub (list) on File_Name; Each with an Index.

  
  Each table has partition-ed Index locally on Job_Activity_Id, STMTDATE, File_Key.

  
  SQL: -.

  merge into out_temp today using

  (

  Select s.trade_id,

  s.notional_amount,

  s.CCY

  of s out_table

  where s.stmtdate = to_date (: b3, 'Dd-MON-yy')

  and s.job_activity_id =: b2

  and s.trade_id <>'VS '.

  and s.file_key =: b1

  )

  yday on

  (

  Today.stmtdate = to_date (: b5, ' DD/MM/YYYY') and today.trade_id =

  yday.trade_id and today.job_activity_id =: b4 and nvl (today.ccy,

  'x') = nvl (yday.ccy, 'x')

  )

  When matched then

  Update

  Set today.amount_tm1 = nvl

  put in correspondence, then setting a day value today.amount_tm1 = NVL (YDAY. NOTIONAL_AMOUNT, 0);

  After being assured, the statistics were gathered, we executed the SQL and found that it was still too much running. Below is the explain plan retrieved by running the wizard to set up the SQL (Plan 1):

  ------------------------------
  Hash value of plan: 3678662366

  -----------------------------------------------------------------------------------------------------------------------------------------
  | ID | Operation | Name                          | Lines | Bytes | Cost (% CPU). Time | Pstart. Pstop |
  -----------------------------------------------------------------------------------------------------------------------------------------
  |   0 | MERGE STATEMENT |                               | 43822.   641K |    22 M (1) | 73:52:39 |       |       |
  |   1.  MERGE                                  | OUT_TEMP |       |       |            |          |       |       |

  
  |   2.   VIEW                                  |                               |       |       |            |          |       |       |
  |*  3 |    TABLE ACCESS BY LOCAL INDEX ROWID | OUT_TABLE |     1.    42.  1781 (1) | 00:00:22 |       |       |
  |   4.     NESTED LOOPS |                               | 43822.  9543K |    22 M (1) | 73:52:39 |       |       |
  |   5.      RANGE OF SINGLE PARTITION |                               | 12447.  2200K |   716 (1) | 00:00:09 |   KEY |   KEY |
  |   6.       PARTITION LIST ALL |                               | 12447.  2200K |   716 (1) | 00:00:09 |     1.    22.
  |*  7 |        TABLE ACCESS BY LOCAL INDEX ROWID | OUT_TEMP | 12447.  2200K |   716 (1) | 00:00:09 |   KEY |   KEY |
  |*  8 |         INDEX RANGE SCAN | IDX_TMP_PART_1 |  8133 |       |   126 (0) | 00:00:02 |   KEY |   KEY |
  |   9.      RANGE OF SINGLE PARTITION |                               |  8852.       |   271 (1) | 00:00:04 |   KEY |   KEY |
  |  10.       SIMPLE LIST OF PARTITION.                               |  8852.       |   271 (1) | 00:00:04 |   KEY |   KEY |
  | * 11 |        INDEX RANGE SCAN | IDX_OUT_TABLE_PART_1 |  8852.       |   271 (1) | 00:00:04 |   KEY |   KEY |
  -----------------------------------------------------------------------------------------------------------------------------------------

  Information of predicates (identified by the operation identity card):
  ---------------------------------------------------

  3 - filter("S".") TRADE_ID ' <>'VS' AND 'Today'. " TRADE_ID "=" S ". "" TRADE_ID "AND
  NVL ('TODAY'. ("' CTL ', 'x') = NVL ("S". "CTL", 'x')) "
  7 - filter("TODAY".") TRADE_ID' <>'VS')
  8 - access("TODAY".") JOB_ACTIVITY_ID ' =: B4 AND 'TODAY'. " STMTDATE "= TO_DATE (: B5,' DD/MM/YYYY)"))
  11 - access("S".") JOB_ACTIVITY_ID "= TO_NUMBER(:B2) AND"S". "STMTDATE"= to_date(:B3,'dd-mon-yy') AND "S". "FILE_KEY"(=:B1) "

  
  
  STA, recommended to accept and apply a profile that has the hash join and the plane of the explain output looks like below (Plan 2):

  --------------------
  Hash value of plan: 1140122102

  ------------------------------------------------------------------------------------------------------------------------------------------------
  | ID | Operation | Name                          | Lines | Bytes | TempSpc | Cost (% CPU). Time | Pstart. Pstop |
  ------------------------------------------------------------------------------------------------------------------------------------------------

  
  |   0 | MERGE STATEMENT |                               | 43822.   641K |       |  2731 (1) | 00:00:33 |       |       |
  |   1.  MERGE                                 | OUT_TEMP |       |       |       |            |          |       |       |
  |   2.   VIEW                                 |                               |       |       |       |            |          |       |       |
  |*  3 |    HASH JOIN |                               | 43822.  9543K |  2352K |  2731 (1) | 00:00:33 |       |       |
  |   4.     RANGE OF SINGLE PARTITION |                               | 12447.  2200K |       |   678 (1) | 00:00:09 |   KEY |   KEY |
  |   3:      PARTITION LIST ALL |                               | 12447.  2200K |       |   678 (1) | 00:00:09 |     1.    22.
  |*  6 |       TABLE ACCESS BY LOCAL INDEX ROWID | OUT_TEMP | 12447.  2200K |       |   678 (1) | 00:00:09 |   KEY |   KEY |
  |*  7 |        INDEX RANGE SCAN | IDX_OUT_TMP_PART_2 |  8133 |       |       |    88 (0) | 00:00:02 |   KEY |   KEY |
  |   8.     RANGE OF SINGLE PARTITION |                               | 60775.  2492K |       |  1782 (1) | 00:00:22 |   KEY |   KEY |
  |   9.      SIMPLE LIST OF PARTITION.                               | 60775.  2492K |       |  1782 (1) | 00:00:22 |   KEY |   KEY |
  | * 10 |       TABLE ACCESS BY LOCAL INDEX ROWID | OUT_TABLE | 60775.  2492K |       |  1782 (1) | 00:00:22 |   KEY |   KEY |
  | * 11 |        INDEX RANGE SCAN | IDX_OUT_TABLE_PART_1 |  8852.       |       |   272 (1) | 00:00:04 |   KEY |   KEY |
  ------------------------------------------------------------------------------------------------------------------------------------------------

  Information of predicates (identified by the operation identity card):
  ---------------------------------------------------

  3 - access("TODAY".") TRADE_ID "=" S ". "" TRADE_ID "AND NVL ('TODAY'. ("' CTL ', 'x') = NVL ("S". "CTL", 'x')) "
  6 - filter("TODAY".") TRADE_ID' <>'VS')
  7 - access("TODAY".") JOB_ACTIVITY_ID ' =: B4 AND 'TODAY'. " STMTDATE "= TO_DATE (: B5,' DD/MM/YYYY)"))
  10 - filter("S".") TRADE_ID' <>'VS')
  11 - access("S".") JOB_ACTIVITY_ID "= TO_NUMBER(:B2) AND"S". "STMTDATE"= to_date(:B3,'dd-mon-yy') AND "S". "FILE_KEY"(=:B1) "

  What I don't understand is, even when the cardinality of 1 Plan is higher than that of Plan 2, why should Oracle opt for a loop nested hash instead of the join? Can someone help me please point to the question and if possible a way to fix it?

  It just occurred to me that we can be deceived by a simple combination of events that links a plan generated when statistics have been exceeded with the STA capture and playback of this plan with statistics that have been updated.  My earlier suggestion on the execution of the query and capture the real plan and the workload (with or without a profile) is the key. In the meantime, however, can you check whether or not collect you statistics immediately after loading your 2 M or more data lines.

  Note: If you only have 2 days of data in the table then the statistics that speak of "yesterday" is 50% out of reach, his stats about two days ago are 100% out of reach and very likely to produce a very stupid and expensive NL plan.

  Concerning

  Jonathan Lewis