Here’s the link to the post: http://blog.tanelpoder.com/2008/12/28/performance-visualization-made-easy-perfsheet-20-beta/

Changelog:

v1.18 12-12-08

• Added INSTANCE_NUMBER and DBID in some views for RAC support
• New view: PHAT$BG_EVENT_SUMMARY
• New view: PHAT$SYS_TIME_MODEL
• New view: PHAT$OSSTAT
• New view: PHAT$EVENT_HISTOGRAM
• New view: PHAT$FILE_HISTOGRAM
• New view: PHAT$TEMP_HISTOGRAM

v1.19 28-04-08

• Added ‘PARTITION BY’ in phat$snapshot to avoid delta values between instance 1 and 2, is now null
• New view: PHAT$UNDOSTAT
• New view: PHAT$ENQUEUE_STATISTICS

Complaints were that a batch was not performing well, and one thing led to another. Queries which were done were of the most simple form:

• select <columns> from <table> where unique_id = <value>
• delete from <table> where unique_id = <value>

Explain plan showed that a full table scan was performed while a unique index existed on the one column which was used as search predicate. Investigation on the size of table and index showed that an index should be used, reading a few block in an index and one block from the table was much more efficient than reading the entire 20+ MB table. So, just to make sure:

• Index rebuild
• Refrefh statistics

No change. Putting an index hint on the queries showed two strange things:

• Cost was lower, why did the optimizer not choose for the index plan ?
• Query used an ‘index full scan’, hello ? It’s a one column index, and were are not doing a count or so.

I decided to make a copy of the table in another schema, and I created a similar index on it. It gave the same behaviour, this did not give much clues. I tried a 10053 trace, despite the fact that I’m not very familiar with it yet. It showed all used variables which the optimizer used, and quite a short analysis of possible queries. It looked like it only calculated one plan, this was strange.

Next experiment I thought of got this problem in a different direction. I decided to export the table to my own test database. Running the export gave this error:

Connected to: Oracle9i Enterprise Edition Release 9.2.0.4.0 - Production
With the Partitioning, OLAP and Oracle Data Mining options
JServer Release 9.2.0.4.0 - Production
Export done in AL32UTF8 character set and UTF8 NCHAR character set

About to export the entire database ...
. exporting tablespace definitions
. exporting profiles
. exporting user definitions
. exporting roles
. exporting resource costs
. exporting rollback segment definitions
. exporting database links
. exporting sequence numbers
. exporting directory aliases
. exporting context namespaces
. exporting foreign function library names
. exporting PUBLIC type synonyms
EXP-00008: ORACLE error 6552 encountered
ORA-06552: PL/SQL: Compilation unit analysis terminated
ORA-06553: PLS-553: character set name is not recognized
EXP-00000: Export terminated unsuccessfully

This looked strange, I checked out invalid objects immediatly. I found 1000+ invalid SYS objects. I tried to fix it:

• utlrp.sql left dozens of invalid KU$ views
• catalog.sql, catproc.sql, utlrp.sql still left them invalid
• compiling an individual KU$ view gave ORA-06552, with a message like ‘type CHAR should be CHAR’
• select on dba_registry showed that catalog was invalid

What has happened ? A ‘grep ORA-’ on the alert log showed the cause: a ‘CREATE DATABASE’ was done on this instance, the environment must have been set wrong when creating a new database. It has run on this existing database, and must have corrupted the catalog. Strange that a rerun of ‘catalog.sql’ did not fix the problem.

Anyway, a restore of the database was decided. After this the queries used the good old ‘index range scan’, and the exp worked again.

I have installed the PERFSTAT schema, and the Phat4Oracle on top of it using the create10g.sql script. No problems found yet. Maybe I will do some more testing later on, but first I have to add some 10g features which are still missing.

9.  New and Changed Features
----------------------------

9.1.  Changes between 10.2  and 11.1

Changes on the Summary Page of the Instance Report
  o  Host
     - Platform name has been added to the Host information.
     - The number of CPU cores and sockets is displayed, where available.
     - Physical Memory is now shown in GB rather than MB.
  o  Snapshot information
     - The DB time and DB CPU in seconds, is now printed close to the
       snapshot Elapsed time.
  o  Load Profile
     - DB time and DB CPU have been added to Load Profile.  Units are
       Per Second, Per Transaction, Per Execute and Per Call
       The addtion of this normalized data assists when examinig
       to reports to see whether the load is comparable.
     - The number of 'Sorts' has been replaced with 'W/A MB processed'.
       Displaying workarea statistics more accurately reflects not
       only sorts, but also other workarea operations such as hash
       joins.  Additionally, using MB processed rather than the number
       of workarea operations indicates the quantity of workarea work
       performed.
     - The following statistics have been removed from the front page,
       as they are no longer considered as important as they once were:
         % Blocks changed per Read
         Recursive Call %
         Rollback per transaction %
         Rows per Sort

  o Instance Efficiency
    - This section has been renamed from 'Instance Efficiency Percentages'
      to 'Instance Efficiency Indicators', as this more accurately
      represents that these values are simply indicators of possible
      areas to consider, rather than conclusive evidence.
   - 'In-memory Sort %' has been replaced with 'Optimal W/A Exec %'
     as the old statistic only showed sorts, and not all workarea
     operations.

Modified sections of the Instance Report
  o  Wait Events and Background Wait Event
     The % of Total Call Time has been added to these sections.  Rows which
     have null for the % of Call Time are Idle Events, and so do not
     contribute to the % Total Call time.

  o  The PGA Memory Advisory now shows the Estimated Time to process the
     data.

  o  The init.ora parameters section now shows the entire parameter value
     rather than truncating the output

  o  SQL sections
     SQL statements which have not completed execution (e.g. have a 0
     execution count as the statement was still running when the End
     snapshot was taken) can now appear in the SQL reports.

New Data captured/reported on - Level 1
  o  Foreground Wait Events
     A new wait event section has been added.  This shows wait events
     for foreground processes only.

  o  OS Statistics - detail (from v$osstat)
     This is a new section which shows the Load, %Busy, %User, %System
     %WIO and %WCPU for each snapshot between the Begin and End snapshot
     Ids specified.

  o  IO Stat by Function - summary (from v$iostat_function)
     This section shows IO statistics broken down by Oracle function.
     A by-function breakdown can help identify where the IO bandwith
     is being consumed (e.g. by RMAN, DBWR, recovery, direct IO).
     The summary section summarizes IO requests and IO rates.

  o  IO Stat by Function - detail (from v$iostat_function)
     Similar to above, however this section shows details how the
     IO requests and IO rates break down by small requests and
     large requests (requests > 128k).

  o  Memory Target Advice (from v$memory_target_advice)
     Shows potential values for memory_target parameter, and
     their estimated cost, in DB time.

  o  Memory Dynamic Components
     Shows infomration about the last resize operation for each
     Memory Component.

  o  Memory Resize Operations
     Shows all known Memory resize operations which occured between
     the Begin and End snapshots.

Obsoleted data
  o  v$temp_histogram view and the corresponding Statspack table have
     been dropped.

• Will there still be a Statspack in 11g ?
• Will it be a ‘little hidden’ like in 10g ? Searching for “statspack install” in the 9R2 manual returns some useful information, this is different in 10R2.
• Will ‘Diagnostics pack’ and ‘Tuning pack’ be free this time ?
• Will there be new tables in the PERFSTAT schema ?
• What columns will be new in PERFSTAT tables ?
• Can I install Phat4Oracle in 11g after creating a PERFSTAT schema ?
• To sum it up: can Phat4Oracle be of any use in 11g ?

$ORACLE_HOME/rdbms/admin/spdoc.txt is one of the first things to check. This text files tells everything about new things which have been added with each new release Oracle release. Besides that, it’s a complete Statspack manual.

• Lot of copy + paste.
• Changing all links.
• Uploading all images.
• Changes reference to new image locations.
• Made thumnails manually, plus link to original image.
• Imported old blog posts.
• Lot of testing to see if all links still work.

New URL also have been put in Phat SQLs, to there’s a new Phat version. No technical improvements though.

Old sites will be shut down soon.

I hope I will find time for some new content on the site, or maybe some technical improvements on Phat.

+ WordPress alows you to have static pages, Blogspot doesn’t. That’s why I have a Google Goup.

+ Better styles with menus for the static pages.

- Can’t upload ZIP’s, PDF’s, etc. I’ve got to find another place for the Phat download.

- Less user friendly.

- Difficult to make proper thumbnails of large images. There’s only one size for a thumbnail.

- Less flexibility in layout of blog postings.

So, all in all: the Google indexing is quite important, and I will migrate the Phat Google group and Blogspot to WordPress. A lot of work already has been done, but there’s still some to do.