Still the definition of HWM is lacking.

High Water Mark defination itself is missing...and required, please provide if you can.

What I understand about HWM (INFORMAL) is that it is a mark
to identify how may blocks had data at least once.

This mark will be changed/refereshed only if

1.) Table is dropped
2.) Table is truncated

or
More data is inserted ( HWM will Increase )

Thanks,
Vivek

excellent explanation!

Thanks, Tom, for 3-step example showing query results before & after Delete and Truncate ...
I believe I finally "got it" on High Water Mark.

HWM was always something that evaded my understanding. Today, ran a search on this site on HWM and now I am HWM aware. Thanks...

In fact, I wanted to ask you all sorts of questions but as I was writing them, I answered all of them... myself...

I wonder if other databases have similar concepts???

Tom,

RDBMS 8.1.6
DB_BLOCK_SIZE 8192
1. I created a table ( 1st extent 1M next 1M  )
2. Inserted 100+ rows
3. Analyzed table
   BLOCKS = 2 EMPTY = 127
4. From DBA_EXTENTS
   BLOCKS = 130
   1 for OS , 1 for segment header, 128 for segments
5. value for '_bump_highwater_mark_count' = 0 , default of 5
6. show_space shows
   freelist = 1 

I expected that 5 blocks are put in the freelist ( default for hwm)
every time Oracle bumps the HWM There is only one on the freelist.
Could you explain for us

Thanks

SQL> create  table t_r (x varchar2(2000)) tablespace RPTSVR
  2  storage
  3  ( initial 1m
  4    next 1m
  5    pctincrease 0
  6  );

Table created.


SQL> begin
  2  for k in 1..111
  3  loop 
  4  insert into t_r values (rpad('*',100,'*'));
  5  end loop;
  6  commit;
  7  end;
  8  /

PL/SQL procedure successfully completed.

SQL> analyze table t_r compute statistics;

Table analyzed.

SQL> select TABLE_NAME,BLOCKS,EMPTY_BLOCKS 
from dba_tables where table_name = 'T_R';

TABLE_NAME                         BLOCKS EMPTY_BLOCKS
------------------------------ ---------- ------------
T_R                                     2          127

SQL> select EXTENT_ID,FILE_ID,BLOCK_ID,BYTES,BLOCKS 
     from dba_extents where SEGMENT_NAME = 'T_R';

 EXTENT_ID    FILE_ID   BLOCK_ID      BYTES     BLOCKS
---------- ---------- ---------- ---------- ----------
         0         19          2    1064960        130

SQL> select KSPPSTVL from x$ksppcv, x$ksppi 
where x$ksppi.INDX = x$ksppcv.INDX and 
ksppinm = '_bump_highwater_mark_count';

KSPPSTVL
----------------------------------------------------
0

SQL> exec show_space('T_R');
Free Blocks.............................1
Total Blocks............................130
Total Bytes.............................1064960
Unused Blocks...........................127
Unused Bytes............................1040384
Last Used Ext FileId....................19
Last Used Ext BlockId...................2
Last Used Block.........................3

PL/SQL procedure successfully completed.

 

SQL>  exec show_space('T_R');
Free Blocks.............................1
Total Blocks............................128
Total Bytes.............................1048576
Unused Blocks...........................122
Unused Bytes............................999424
Last Used Ext FileId....................5
Last Used Ext BlockId...................6881
Last Used Block.........................6

PL/SQL procedure successfully completed.

SQL>  insert into t_r values (rpad('*',1000,'*'));

1 row created.

SQL> commit;

Commit complete.

SQL>  exec show_space('T_R');
Free Blocks.............................5
Total Blocks............................128
Total Bytes.............................1048576
Unused Blocks...........................117
Unused Bytes............................958464
Last Used Ext FileId....................5
Last Used Ext BlockId...................6881
Last Used Block.........................11

PL/SQL procedure successfully completed.

SQL> select count(*) from t_r;

  COUNT(*)
----------
        36

It seems that the first 5 blocks Oracle allocates 1 block 
at a time and subsequent to first 5, bumps 5 blocks in the
freelist for HWM

Thanks 

Ummm, disagree:

SQL>  exec show_space('T_R');
Free Blocks.............................1
Total Blocks............................128
Total Bytes.............................1048576
Unused Blocks...........................122


128 blocks - 122 Unused blocks = 6 blocks "used".  1 block is for the segment header which means there were 5 blocks in the table.  You filled up 4 of all of the way with data and only filled part of the 5th block.

It was *always* 5 blocks. 

One of my packages truncates a table “A” once per every 2 hours and then inserts and updates the table with the required data.

1. Truncate table A;

2. Insert into table A;

3. Update table A:

4. No delete statements.

The current size of the table is 340 MB. When I create a copy of this table using

Create table B nologging as select * from A;

Then the target table “B” size is around 24 MB.

Why my original table size was 15 times higher then actual size? How to fix this problem?

Please excuse me if it’s a basic question.

The original storage parameters of A were such that the initial, next, pctincrease, minextents said "Table A should allocate 340 mb"

The storage parameters of B are different.  Consider:

ops$tkyte@ORA817DEV> create table a ( x int ) tablespace users storage ( initial 340m );
Table created.

ops$tkyte@ORA817DEV> select sum(bytes) from user_extents where segment_name = 'A';

SUM(BYTES)
----------
 357040128

ops$tkyte@ORA817DEV> create table b tablespace users as select * from a;
Table created.

  1* select sum(bytes) from user_extents where segment_name = 'B'
ops$tkyte@ORA817DEV> /

SUM(BYTES)
----------
    524288


so, look at A's initial, next, pctincrease, minextents -- that'll tell you what truncate will take it back to.


 

Thanks for your quick answer.

I verified the initial extents for both Table A and B and it was 65536.

Here are the DDL’s for my tables

1.
CREATE TABLE A (
Col1 NUMBER (18),
Col2 NUMBER (18),
Col3 NUMBER (18),
Col4 VARCHAR2 (4000),
Col5 VARCHAR2 (4000),
... –other columns
...
)
TABLESPACE TOOLS NOLOGGING
PCTFREE 10
PCTUSED 40
INITRANS 1
MAXTRANS 255
STORAGE (
INITIAL 65536
NEXT 65536
PCTINCREASE 0
MINEXTENTS 1
MAXEXTENTS 10000
FREELISTS 1 FREELIST GROUPS 1 )
NOCACHE;

2.
CREATE TABLE B (
Col1 NUMBER (18),
Col2 NUMBER (18),
Col3 NUMBER (18),
Col4 VARCHAR2 (4000),
Col5 VARCHAR2 (4000),
... –other columns
...
)
TABLESPACE TOOLS NOLOGGING
PCTFREE 10
PCTUSED 40
INITRANS 1
MAXTRANS 255
STORAGE (
INITIAL 65536
NEXT 65536
PCTINCREASE 0
MINEXTENTS 1
MAXEXTENTS 10000
FREELISTS 1 FREELIST GROUPS 1 )
NOCACHE;

I am still not sure why table A used 340 mb and table B used 24 MB.

Hi Tom, 

I'm pretty new to the DBA role, so excuse if this question is obvious, but i'm a bit perplexed by the results of show_space for this setup. I need to know what you recommend to change for this (relatively old) application storage setup.

SQL> CREATE TABLESPACE TEST DATAFILE 'C:\TEMP\TEST.DBF' SIZE 100M REUSE autoextend on EXTENT MANAGEMENT LOCAL
  2        UNIFORM
  3        size 100k ONLINE ;

Tablespace created.

SQL> CREATE TABLE TEMP ( A VARCHAR2(10) ) TABLESPACE TEST;

Table created.

Okay, I've created the table. The HWM (as I expect) is 1.

SQL> exec show_space('TEMP');
Shows segment information.
Tablespace..................................................TEST
Tablespace Block Size.......................................8192
Num Extents.................................................1
Total Blocks................................................13
Total Bytes.................................................106496
Last Used Block (HWM).......................................1
Unused Blocks (blocks above the HWM)........................12
Unused Bytes................................................98304
Free Blocks (blocks on the freelist)........................0

PL/SQL procedure successfully completed.

If I now populate with loads of data...

SQL> BEGIN
  2    FOR i IN 1..100000 LOOP
  3      INSERT INTO temp VALUES ('XXXXXXXXXX');
  4    END LOOP;
  5  END;
  6  /

PL/SQL procedure successfully completed.

SQL> EXEC SHOW_SPACE('TEMP');
Shows segment information.
Tablespace..................................................TEST
Tablespace Block Size.......................................8192
Num Extents.................................................18
Total Blocks................................................234
Total Bytes.................................................1916928
Last Used Block (HWM).......................................5
Unused Blocks (blocks above the HWM)........................8
Unused Bytes................................................65536
Free Blocks (blocks on the freelist)........................5

PL/SQL procedure successfully completed.

Can you explain to me why the total blocks for the segment has gone to 234, but the HWM is still only 5 (or am I reading this wrong)?

Thanks for all your invaluable help.
 

Very informative

We have been reading about HWM, but why does oracle maintain a HWM which is not the number of blocks currently used. Why does it need it to know the number of blocks which were ever used by a table..?

Hi,
The blocks are not currently in use by that table, The data has been deleted and also been committed.
Hence while scanning we should scan only the blocks which currently have data, not necessarily had data at one point of time,
Still can't comprehend why are we scanning the blocks which are not currently in use. can you please elaborate


Thanks

Hi tom,
I still dont get it, I can understand you getting mad at me, as u have said its a complex structure not a lifo/fifo methodoly. I can understand the concept if i can understand why it needs to do that.

If the data has not been committed i can understand the blocks being read, but if the data has been commited, only those should be read, as others are not in use.
if we take just 2 session,
1) Delete * from tablename <table1>
Deletes blocks from 1-25 (hypotetical)

2) Insert into the tablename <table1> select * from table2
This would insert from blocks 26 to 50 (hypotetical).

we have a third session which is doing a
select * from <table1>
if the first delete has not been commited it and insert has been it needs to scan blocks 1 to 50, but if first delete has been commited then it should just read blocks 26 to 50 as they are the one in use.

Can you plesase explain a little more in detail.

Thanks for your reply

Hi Tom,
We do not need to look at eack row, but the blocks can be read based on the freelist,I am not trying to give a solution for HWM, but i need to know why do we ever have something like HWM,what overhead does it reduce and what are the advantages, as uptill now i hv read more of disadvantages of HWM, can you point me to any documentation which would gove details about the comple table structure and why HWM is used.
Since most of my queries have always been answered by you, thats whay i am invariably depending on you. :)


Thanks in Advance

Hi Tom,

Why do so many people worry about HWM? Is it just for mental exercise to satisfy one's curiosity or does knowing it have more practical use than I am aware of? All I care to know is that full table scan will scan till HWM. I don't recall ever have to manipulate HWM.

"c) this becomes most obvious to people running the RBO who like to constantly
issue "select count(*) from t" just to see how many rows this table might
contain". They will constantly full scan upto the HWM -- they should use the
CBO which will index full/fast full scan instead -- removing this problem (or
just stop counting the rows!) "

How is the index full/fast full scan better? examples please.

Hi Tom, Rather John

"Why do so many people worry about HWM? Is it just for mental exercise to satisfy
one's curiosity or does knowing it have more practical use than I am aware of?
All I care to know is that full table scan will scan till HWM. I don't recall
ever have to manipulate HWM. "
The full table scan you must be performing for small rows
try performing Full table scans for tables with millions of Rows, When it comes to tunning a Database i think even a few seconds make a difference.
"Yes its faster then before the data does come up quickly now", A lot of user will give you this reaction even if the data just comes quickly by a few seconds, and as a proper learning process, i like to know why is it being done so that we can try and exploit it in a better way, rather then half of knowledge.

More over what will be ur reaction will be when you try to explain BA's and Project managers that its doing a Full table scan and the insert statement is writting above HWM as we have used a Append hint to increase the performance of upload.All they will ask is Why is HWM maintained, what do you do? you raise your two hands in air and say dont know ask oracle. I certainly dont like to do that.

Thanks tom for the info

Tom there is a script on www.oracle-base.com
used to rebuild indexes it talks about a high water mark
on tablespaces..

"Rebuilding indexes within the tablespace raises the high water mark on the tablespace. This chews disk space which cannot be easily recovered. Unix)."

I know about table HWM but what the heck is a tablspace
HWM and what are it's impacts

Thanks very much Tom
for the clear explainantion on Tablespace HWM

i will search for index rebuilds now as you suggested

Kind Regards Trevor

I have a table which does some constant deletes and inserts, as this is the table we store the ids to which the users have access based on their profile,like user x has access to all ids or all of these 10,000 ids, We need to populate the table everytime the user logds in, Delete on this table is taking quite some time as the number of rows per user go upto .4 million.
Is there a way to delete without redo or speed up the insert and delete, i am currently using bulk insert, is bulk delete faster, i cant use truncate as the other users maybe using the table, dont want to partiion and truncate as the user creation is thru an application. Any ideas.

its a web application wher the user does not necessary get that account and it has connection pooling implemented in app server to ensur the connections are used up rationally

Tom,

In the output of the show_space procedure, does the value Total Blocks = Last Used Block + Unused Block?

How do i interpret the following output?

'ATTRIBUTE'
Free Blocks.............................3
Total Blocks............................18432
Total Bytes.............................150994944
Unused Blocks...........................284
Unused Bytes............................2326528
Last Used Ext FileId....................11
Last Used Ext BlockId...................202633
Last Used Block.........................740

No -- since an object could be in extents all over the place (in many files).



free blocks -- blocks on the freelist.

total blocks/bytes -- blocks/bytes allocated to the segment

unused blocks/bytes -- blocks/bytes allocated to the segment but not yet used to store anything ever (eg: unformatted, above the high water mark)

last used ext file/block id -- a pointer to the front of the "last extent" of the table.  Consider:

ops$tkyte@ORA817DEV> create table t ( x varchar2(4000) ) tablespace system storage ( initial 32k next 32k pctincrease 0 );
 
Table created.
 
ops$tkyte@ORA817DEV>
ops$tkyte@ORA817DEV> exec show_space( 'T' )
Last Used Ext FileId....................1
Last Used Ext BlockId...................30858
Last Used Block.........................1
 
PL/SQL procedure successfully completed.
 
ops$tkyte@ORA817DEV> select block_id first_block, block_id+blocks-1 last_block from dba_extents where segment_name = 'T' and owner = user;
 
FIRST_BLOCK LAST_BLOCK
----------- ----------
      30858      30861

 
ops$tkyte@ORA817DEV>
ops$tkyte@ORA817DEV> insert into t select rpad('*',4000,'*') from all_objects where rownum <= 5;
 
5 rows created.
 
ops$tkyte@ORA817DEV> exec show_space( 'T' )
Last Used Ext FileId....................1
Last Used Ext BlockId...................30867
Last Used Block.........................4
 
PL/SQL procedure successfully completed.
 
ops$tkyte@ORA817DEV> select block_id first_block, block_id+blocks-1 last_block from dba_extents where segment_name = 'T' and owner = user;
 
FIRST_BLOCK LAST_BLOCK
----------- ----------
      30858      30861
      30867      30870

 
ops$tkyte@ORA817DEV> insert into t select rpad('*',4000,'*') from all_objects where rownum <= 5;
 
5 rows created.
 
ops$tkyte@ORA817DEV> exec show_space( 'T' )
Last Used Ext FileId....................1
Last Used Ext BlockId...................19841
Last Used Block.........................5
 
PL/SQL procedure successfully completed.
 
ops$tkyte@ORA817DEV> select block_id first_block, block_id+blocks-1 last_block from dba_extents where segment_name = 'T' and owner = user;
 
FIRST_BLOCK LAST_BLOCK
----------- ----------
      30858      30861
      30867      30870
      19841      19848

ops$tkyte@ORA817DEV> insert into t select rpad('*',4000,'*') from all_objects where rownum <= 5;
 
5 rows created.
 
ops$tkyte@ORA817DEV> exec show_space( 'T' )
Last Used Ext FileId....................1
Last Used Ext BlockId...................19841
Last Used Block.........................8
 
PL/SQL procedure successfully completed.
 
ops$tkyte@ORA817DEV> select block_id first_block, block_id+blocks-1 last_block from dba_extents where segment_name = 'T' and owner = user;
 
FIRST_BLOCK LAST_BLOCK
----------- ----------
      30858      30861
      30867      30870
      19841      19848
 
ops$tkyte@ORA817DEV> insert into t select rpad('*',4000,'*') from all_objects where rownum <= 5;
 
5 rows created.
 
ops$tkyte@ORA817DEV> exec show_space( 'T' )
Last Used Ext FileId....................1
Last Used Ext BlockId...................30875
Last Used Block.........................4
 
PL/SQL procedure successfully completed.
 
ops$tkyte@ORA817DEV> select block_id first_block, block_id+blocks-1 last_block from dba_extents where segment_name = 'T' and owner = user;
 
FIRST_BLOCK LAST_BLOCK
----------- ----------
      30858      30861
      30867      30870
      19841      19848
      30871      30874
      30875      30878
 
ops$tkyte@ORA817DEV> insert into t select rpad('*',4000,'*') from all_objects where rownum <= 5;
 
5 rows created.
 
ops$tkyte@ORA817DEV> exec show_space( 'T' )
Last Used Ext FileId....................1
Last Used Ext BlockId...................30879
Last Used Block.........................4
 
PL/SQL procedure successfully completed.
 
ops$tkyte@ORA817DEV> select block_id first_block, block_id+blocks-1 last_block from dba_extents where segment_name = 'T' and owner = user;
 
FIRST_BLOCK LAST_BLOCK
----------- ----------
      30858      30861
      30867      30870
      19841      19848
      30871      30874
      30875      30878
      30879      30882
 
6 rows selected.
 


<b>It'll point to the "last" extent.</b>
 

When oracle adjusts HWM position, does it lock the table? Thanks.

From, </code> http://download-west.oracle.com/docs/cd/B14117_01/server.101/b10739/schema.htm#sthref1783 <code>

<quote>As with other DDL operations, segment shrink causes subsequent SQL statements to be reparsed because of invalidation of cursors unless you specify the COMPACT clause,<quote>

In 10g, when the segment is shrunk by deallocating space below hwm, it locks the table. right? that means it locks the table when it moves hwm. Therefore, does oracle lock the table when it adjusts HWM? maybe I am confused! Any comments? Thanks.

Refering David's question above, does oracle lock the table when segment shrink operation is done? I would be surprised if it did. Thanks.

Tom,

1) How do we find out if a given table has a high water mark?
2) The way to reset a table with high water mark is a)exp/trunc/imp b) alter table t move, rebuild index. Are they any other ways of doing this?

DB version - 9.2.0.6

Thank you

>>1) find me a real table without one! they all have high water marks...
That's correct, but isn't a high water mark relative? Ex. If a table has 20 million records and if we delete 17 million records, then a select count(*) would take a relatively long time reading 20 million records to 3 million records. I am of the assumption that if the high water mark is really high, them the time taken to do selects goes up and so the performance is impacted. Please let me know if my assumtion is wrong?
If my assumption is right, I wanted to find tables with really high water mark and fix them - and so the request to find out tables with high/really high water mark.


Thank you

Hi Tom,
why exactly do we need a High Water Mark.Who sets the High Water Mark and why?

Hi Tom,
If I have a huge table, I dont want to use analyze or dbms_stats to get the HWM. But I must be stored somewhere so oracle knows where to stop when doing full table scan.

(1)
Just curious where HWM information is stored, so I can get it without consuming lots of resource to do analyze or dbms_stats. I guess its in segment header and dbms_space package reads segment header, right?

(2)
If it's in segment header, then how could i interpret the data of the block dump in order to find HWM?

Thanks.

Hi Tom,
I was not asking how to get HWM, I am not sure whether my questions were clear enough. There were simply two questions
(1) Where is HWM information stored, is it in segment header so dbms_space package can read it.
(2) How to interpret the direct segment header block dump to find out HWM if it's not a business secret.

Thank you as always

Tom,

How does one go about resetting the HWM for a table and a datafile? I have found so many tables that have a few thousand rows that initially had millions of rows.

Thanks

Referring to your comment "alter table t move - to offline reorg"

When I use this command, I see that the 'blocks' column in the view user_tables is reset to null. Only when I analyze again, the actual blocks occupied is shown that is much less than before reorg. That means the table has been reorg'ed. So I did not understand your comment to use this as offline reorg.

Hi Tom

Is there any way to get the list of tables were not used for sometime?
I don't have auditing on table

Thanks
Ram

You potentially could keep snapshots of v$segment_statistics somewhere. If a particular object never gets any statistics then it may have never been touched. Though there may always be a chance the stats somehow were aged out or flushed before each snapshot was taken...
Then again you could rename the table and see if anybody eventually yells about it, and hope it's not one of those "it's important we see this data once every 5 years sort of thing" :-)

Tom,

Is there a formula to knows if a table need to reset the HWM ( alter table t move )?

In 10g you also have the option to shrink. See
</code> https://asktom.oracle.com/Misc/oramag/on-favorites-and-connect-by.html <code>
I think the advisor still leaves it to you to decide whether to go ahead and shrink or not.

hwm

The "alter table .... shrink space cascade" seems a bit slow to me. I created a test table with 100,000 blocks, deleted all but 100 out of 100,000 rows. Then I ran the shrink, which took 994 seconds (947 in another database). For comparison, I did an "alter table ... move", this took 175 seconds. Is this typical?
Also, it's a pity there's no info in v$session_longops when running the shrink...

Oracle Database 10g Enterprise Edition Release 10.1.0.5.0 - Production
With the Partitioning, Real Application Clusters and Data Mining options
System name: Linux
Release: 2.4.21-32.0.1.ELhugemem
Version: #1 SMP Tue May 17 17:43:22 EDT 2005
Machine: i686

FYI

Thanks for all the useful information on this link.

I am trying to migrate from Oracle8.1.7 to 10g. I created a test DB and am trying to import the structure from the production DB without the data.
The production DB runs into tera bytes and all I need right now is to get the structure in there so that I can check all PL/SQL and compile forms and reports.
But when I try to import, the tablespaces start growing to (possibly) accomodate the data in each table per statistics that get inported and I fall short on space on the new system (since this is only a test system and I am limited to 100 gigs).
After my import fails, I tried everything from shrinking tables to shrinking indexes to re-analyzing tables and truncating tables and coalescing tablespaces. All tablespaces now show as empty and all data files are also almost empty. But it is not letting me re-size any of the data files.
What do I do?

Hi Tom,

An update on the issue above. I just found out that for some reason, when I am running my import, the tables that are getting created have initial sizes running into 100's of MB's for all large tables. That is probably why after doing all trciks I am not able to reclaim the space.
How do I resize the tables to smaller size?
If I use 'create table as...' then not only do I not know how to give size definition using that syntax, but also I lose all index/contraint information which I would have to re-create separately.

Hi Tom,

I managed to fix that problem by:

alter table T move
storage ( initial 1k next 1k minexents 1 pctincrease 0 );

and

alter index I rebuild
storage ( initial 1k next 1k minexents 1 pctincrease 0 );

Then I executed your 'maxshrink' script and was able to regain about 40 GB of space.

Hello Tom,

I saw in this thread that full scan is reading table up to the HWM and therefore I have a question.
I created a table:
OPS$SHMYG@REX> create table test (customer_id number);

Then I put there some data:
OPS$SHMYG@REX> begin
for i in 1..100000
loop
insert into test values (i);
end loop;
end;
/

Now the storage consumed by the table looks as follows:
OPS$SHMYG@REX> @space TEST
Free Blocks.............................
Total Blocks............................256
Total Bytes.............................2097152
Unused Blocks...........................91
Unused Bytes............................745472
Last Used Ext FileId....................5
Last Used Ext BlockId...................20617
Last Used Block.........................37

So, from my understanding, HWM is 165 blocks.
But when I run a simple query:
OPS$SHMYG@REX> select * from test;

and tkprof it, I see the following:

select *
from
test


call count cpu elapsed disk query current rows
------- ------ -------- ---------- ---------- ---------- ---------- ----------
Parse 1 0.00 0.00 0 0 0 0
Execute 1 0.00 0.00 0 0 0 0
Fetch 101 0.04 0.05 0 258 0 100000
------- ------ -------- ---------- ---------- ---------- ---------- ----------
total 103 0.04 0.05 0 258 0 100000

Misses in library cache during parse: 0
Optimizer mode: ALL_ROWS
Parsing user id: 60

Rows Row Source Operation
------- ---------------------------------------------------
100000 TABLE ACCESS FULL TEST (cr=258 pr=0 pw=0 time=300117 us)

So the question is: why do I have 258 blocks read while (from my understanding) it should be 165 (as HWM implies)?

Thanks in advance

Hi Tom,

I installed the show_space procedure and ran it for two tables in one of the schema.

Here is the results I got,

SQL>

Free Blocks.............................
Total Blocks............................4096
Total Bytes.............................67108864
Total MBytes............................64
Unused Blocks...........................0
Unused Bytes............................0
Last Used Ext FileId....................27
Last Used Ext BlockId...................58564
Last Used Block.........................64

PL/SQL procedure successfully completed

Free Blocks.............................
Total Blocks............................4608
Total Bytes.............................75497472
Total MBytes............................72
Unused Blocks...........................509
Unused Bytes............................8339456
Last Used Ext FileId....................27
Last Used Ext BlockId...................276740
Last Used Block.........................3

PL/SQL procedure successfully completed

I was thinking, the total Blocks should be the same as Last Used Block(HWM) + Unused Block(Blocks above HWM) + Free Blocks (If there are any empty block below the HWM).

I think we have LMT tablespaces and on Oracle 9205.

Is my understanding is correct?? or am I missing something or doesn't understand something??

Thanks for your time and inputs,

Kandy Train

Hi Tom,

Okay, I read through a little bit and understand the equation total blocks = free blocks + unused blocks + last used block is not true.

But in my two examples, My total blocks are so huge and the HWM is very low. How would I claim those allocated blocks to the segment??

I am planning to move the table and rebuild the indexes, that should reset the HWM, but not sure how I can deallocate some of those blocks from 'Total Blocks' as it looks like a waste leave such space..

Thanks,

Kandy Train

Hi Tom,

How could you make up the total blocks in the show_space program. My problem is, total blocks are so huge, and I cannot add unused blocks + HWM + Free Blocks to make up the total blocks.

I would like to understand where all the blocks are that make up the total blocks, and how many of them really used and how many of them really free(Empty).

Thanks,

Kandy Train

Hi Tom,

Thank you for the explanation and Yes, I now understand that the Last Used Block is a relating no and is in the last used extent. That's not the HWM. But If I know the
"Last Used Extent", then as in your example, I can count the HWM correctly.

But how do I know "Last Used Extent" ??

This is a small example I have based my questions on,

select segment_name, count(*) num_of_extents, sum(blocks) total_blocks
from dba_extents
where owner = 'SCOTT'
and segment_name = 'INVENTORY_PART_TAB'
group by segment_name;

set serveroutput on
exec show_space('INVENTORY_PART_TAB', 'SCOTT', 'TABLE');

--Results
SQL>

SEGMENT_NAME NUM_OF_EXTENTS TOTAL_BLOCKS
-------------------------------------------------------------------------------- -------------- ------------
INVENTORY_PART_TAB 65 4608

Free Blocks.............................
Total Blocks............................4608
Total Bytes.............................75497472
Total MBytes............................72
Unused Blocks...........................509
Unused Bytes............................8339456
Last Used Ext FileId....................13
Last Used Ext BlockId...................292164
Last Used Block.........................3

PL/SQL procedure successfully completed

Hi,

I am having a table in 9.2 database in a dictonary managed tablespace, after collecting statistics it is showing following

0:57:43 SQL> select table_name, initial_extent, num_rows, blocks, empty_blocks, avg_row_len
0:57:52 2 from user_tables
0:57:52 3 where table_name='TWW11';

ABLE_NAME INITIAL_EXTENT NUM_ROWS BLOCKS EMPTY_BLOCKS AVG_ROW_LEN
-------------- -------------- ---------- ---------- ------------ -----------
TWW11 10485760 37310 387404 0 231

My block size is 4k. It means for 8m ( 37310*231 ) table, total size was occupied was 1.5 gb.

empty_blocks are 0 it means there are no blocks allocated over high water mark. initial extent size is 10m.

If I try to deallocate, it is not working ( it should not work if empty blocks are 0 )

12:02:33 SQL> alter table tww11 deallocate unused;

Table altered.

12:02:51 SQL> exec dbms_stats.gather_table_stats('A', 'TWW11');

PL/SQL procedure successfully completed.

12:05:08 SQL> select table_name, initial_extent, num_rows, blocks, empty_blocks, avg_row_len
12:05:12 2 from user_tables
12:05:12 3 where table_name='TWW11';

TABLE_NAME INITIAL_EXTENT NUM_ROWS BLOCKS EMPTY_BLOCKS AVG_ROW_LEN
--------------- -------------- ---------- ---------- ------------ -----------
TWW11 10485760 37310 387404 0 231

Ultimately, I have to use alter table move clause to reduce a no of blocks allocated

12:05:12 SQL> alter table tww11 move storage (initial 50k next 50k pctincrease 0 minextents 1);

Table altered.

12:09:03 SQL> exec dbms_stats.gather_table_stats('A', 'TWW11');

PL/SQL procedure successfully completed.

12:10:40 SQL> select table_name, initial_extent, num_rows, blocks, empty_blocks, avg_row_len
12:11:10 2 from user_tables
12:11:10 3 where table_name='TWW11';

TABLE_NAME INITIAL_EXTENT NUM_ROWS BLOCKS EMPTY_BLOCKS AVG_ROW_LEN
--------------- -------------- ---------- ---------- ------------ -----------
TWW11 53248 37310 2561 0 231

My questions is why some blocks are not showing under empty_blocks ? Why it has been allocated so many blocks for such a small table size ?

SQL> select initial_extent, next_extent, pct_increase from dba_tablespaces where tablespace_name='A_D01';

INITIAL_EXTENT NEXT_EXTENT PCT_INCREASE
-------------- ----------- ------------
102400 12288 0

Pl tell me.

thanks & regards
pjp

Hi Tom,

If my table size 8k, total allocated size is 1.5gb, why some of blocks are not showing under empty blocks ? I am using 9.2 database and dictionary managed tablespace.

Kindly suggest.

thanks & regards
pjp

Hi Tom,

Thanks for your explanation of empty blocks. I agree with this. But when you delete the data, Oracle should show me blocks under freelist, why it is not showing me

1 select table_name, blocks, empty_blocks, num_freelist_blocks, pct_free, pct_used, num_rows,
2 avg_row_len
3 from user_tables
4* where table_name = 'TWW44'
10:40:08 SQL> /

TABLE_NAME BLOCKS EMPTY_BLOCKS NUM_FREELIST_BLOCKS PCT_FREE PCT_USED NUM_ROWS AVG_ROW_LEN
---------- ---------- ------------ ------------------- ---------- ---------- ---------- -----------
TWW44 440917 0 0 10 40 36997 231

What is a reason it is not showing under freelist ?

thanks & regards
pjp

Hi Tom

We have Table News_story.

CREATE TABLE NEWS_STORY
(
STORY_TS DATE NOT NULL,
NEWS_STORY_ID NUMBER,
STORY_TM_CT NUMBER NOT NULL,
NEWS_SRC_NM VARCHAR2(30 BYTE) NOT NULL,
WIRE_SRC_CD VARCHAR2(4 BYTE) NOT NULL,
STORY_HDLN_TX VARCHAR2(160 BYTE) NOT NULL,
STORY_WORD_CT NUMBER(6) NOT NULL,
STORY_BODY_BK CLOB NOT NULL,
SM_STORY_FL CHAR(1 BYTE) NOT NULL,
DOC_TYPE_CD NUMBER,
VISTA_NEWS_ID VARCHAR2(50 BYTE),
LOAD_TS DATE
)
TABLESPACE SONAR_IDX_TS
PCTUSED 50
PCTFREE 10
INITRANS 1
MAXTRANS 255
STORAGE (
INITIAL 400M
NEXT 200M
MINEXTENTS 1
MAXEXTENTS 500
PCTINCREASE 0
)
LOGGING
PARTITION BY RANGE (STORY_TS)
(
PARTITION NEWS_STORY_P200609 VALUES LESS THAN (TO_DATE(' 2006-10-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIAN'))
LOGGING
NOCOMPRESS
TABLESPACE SONAR_IDX_TS
LOB (STORY_BODY_BK) STORE AS
( TABLESPACE CLOB_TS
ENABLE STORAGE IN ROW
CHUNK 16384
PCTVERSION 10
NOCACHE
STORAGE (
INITIAL 200M
NEXT 347304K
MINEXTENTS 4
MAXEXTENTS 2147483645
PCTINCREASE 1
FREELISTS 1
FREELIST GROUPS 1
BUFFER_POOL DEFAULT
)
)
PCTUSED 50
PCTFREE 10
INITRANS 1
MAXTRANS 255
STORAGE (
INITIAL 400M
NEXT 200M
MINEXTENTS 1
MAXEXTENTS 500
PCTINCREASE 0
FREELISTS 1
FREELIST GROUPS 1
BUFFER_POOL DEFAULT
),
PARTITION NEWS_STORY_P200610 VALUES LESS THAN (TO_DATE(' 2006-11-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIAN'))
LOGGING
NOCOMPRESS
TABLESPACE SONAR_IDX_TS
LOB (STORY_BODY_BK) STORE AS
( TABLESPACE CLOB_TS
ENABLE STORAGE IN ROW
CHUNK 16384
PCTVERSION 10
NOCACHE
STORAGE (
INITIAL 200M
NEXT 330424K
MINEXTENTS 4
MAXEXTENTS 2147483645
PCTINCREASE 1
FREELISTS 1
FREELIST GROUPS 1
BUFFER_POOL DEFAULT
)
)
PCTUSED 50
PCTFREE 10
INITRANS 1
MAXTRANS 255
STORAGE (
INITIAL 400M
NEXT 200M
MINEXTENTS 1
MAXEXTENTS 500
PCTINCREASE 0
FREELISTS 1
FREELIST GROUPS 1
BUFFER_POOL DEFAULT
),

.......

Clob fied is store in store in seperate tablespace clob_ts. This table is occupying around 600 GB of . CLOB data occupies more than 500 GB. So we of thought of getting rid of data prior to 2005 and get back some space. What we did was create a staging table copy all the data to the staging table (other than clob field) ,truncate the partition and then exchange the partition by staging table. In this way we got back around 230 GB of space. But this space we were not able to allocate to other tablespace as freed space is below HWM.

1. How to reset HWM?
2. How to organize the data? We can not export and import as we dont have that much of space.
3. Clob_Ts Tablespace is already around 400 GB, rest of the tables are under different tablespace whose size is less than 100 GB. Do you see any possibility of performance impact if we move tables from other tablespace to clob_ts.


Thanks
Koshal

TABLESPACE_NAME FILE_NAME HIGHWATER
CLOB_TS /DB07/SONARP/ORACLE/data01/SONAR_CLOB_TS_01.dbf 8360517632
CLOB_TS /DB07/SONARP/ORACLE/data01/SONAR_CLOB_TS_07.dbf 8319639552
CLOB_TS /DB07/SONARP/ORACLE/data01/SONAR_CLOB_TS_13.dbf 8360517632
CLOB_TS /DB07/SONARP/ORACLE/data01/SONAR_CLOB_TS_19.dbf 8360517632
CLOB_TS /DB07/SONARP/ORACLE/data01/SONAR_CLOB_TS_29.dbf 4208156672
CLOB_TS /DB07/SONARP/ORACLE/data02/SONAR_CLOB_TS_02.dbf 16749125632
CLOB_TS /DB07/SONARP/ORACLE/data02/SONAR_CLOB_TS_08.dbf 8361582592
CLOB_TS /DB07/SONARP/ORACLE/data02/SONAR_CLOB_TS_14.dbf 8360517632
CLOB_TS /DB07/SONARP/ORACLE/data02/SONAR_CLOB_TS_20.dbf 10457669632
CLOB_TS /DB07/SONARP/ORACLE/data02/SONAR_CLOB_TS_55.dbf 1553334272
CLOB_TS /DB07/SONARP/ORACLE/data03/SONAR_CLOB_TS_21.dbf 16707182592
CLOB_TS /DB07/SONARP/ORACLE/data04/SONAR_CLOB_TS_04.dbf 8361582592
CLOB_TS /DB07/SONARP/ORACLE/data04/SONAR_CLOB_TS_10.dbf 8361582592
CLOB_TS /DB07/SONARP/ORACLE/data04/SONAR_CLOB_TS_16.dbf 8366907392
CLOB_TS /DB07/SONARP/ORACLE/data04/SONAR_CLOB_TS_22.dbf 8318574592
CLOB_TS /DB07/SONARP/ORACLE/data04/SONAR_CLOB_TS_25.dbf 7815258112
CLOB_TS /DB07/SONARP/ORACLE/data04/SONAR_CLOB_TS_28.dbf 10919043072
CLOB_TS /DB07/SONARP/ORACLE/data05/SONAR_CLOB_TS_05.dbf 8361582592
CLOB_TS /DB07/SONARP/ORACLE/data05/SONAR_CLOB_TS_11.dbf 8318574592
CLOB_TS /DB07/SONARP/ORACLE/data05/SONAR_CLOB_TS_17.dbf 8360517632
CLOB_TS /DB07/SONARP/ORACLE/data05/SONAR_CLOB_TS_23.dbf 8360517632
CLOB_TS /DB07/SONARP/ORACLE/data05/SONAR_CLOB_TS_27.dbf 1901289472
CLOB_TS /DB07/SONARP/ORACLE/data06/SONAR_CLOB_TS_06.dbf 8319639552
CLOB_TS /DB07/SONARP/ORACLE/data06/SONAR_CLOB_TS_12.dbf 8360517632
CLOB_TS /DB07/SONARP/ORACLE/data06/SONAR_CLOB_TS_18.dbf 8360517632
CLOB_TS /DB07/SONARP/ORACLE/data06/SONAR_CLOB_TS_24.dbf 4166213632
CLOB_TS /DB07/SONARP/ORACLE/data06/SONAR_CLOB_TS_30.dbf 8192745472
CLOB_TS /DB07/SONARP/ORACLE/data06/SONAR_CLOB_TS_31.dbf 8192745472
CLOB_TS /DB07/SONARP/ORACLE/data07/SONAR_CLOB_TS_03.dbf 8360517632
CLOB_TS /DB07/SONARP/ORACLE/data07/SONAR_CLOB_TS_09.dbf 8319639552
CLOB_TS /DB07/SONARP/ORACLE/data07/SONAR_CLOB_TS_15.dbf 8360517632
CLOB_TS /DB07/SONARP/ORACLE/data07/SONAR_CLOB_TS_26.dbf 8360517632
CLOB_TS /DB07/SONARP/ORACLE/data07/SONAR_CLOB_TS_40.dbf 20927201280
CLOB_TS /DB07/SONARP/ORACLE/data07/SONAR_CLOB_TS_41.dbf 15681708032
CLOB_TS /DB07/SONARP/ORACLE/data08/SONAR_CLOB_TS_32.dbf 20775657472
CLOB_TS /DB07/SONARP/ORACLE/data08/SONAR_CLOB_TS_33.dbf 20775657472
CLOB_TS /DB07/SONARP/ORACLE/data08/SONAR_CLOB_TS_36.dbf 6095593472
CLOB_TS /DB07/SONARP/ORACLE/data09/SONAR_CLOB_TS_27.dbf 6286303232
CLOB_TS /DB07/SONARP/ORACLE/data09/SONAR_CLOB_TS_37.dbf 20844437504
CLOB_TS /DB07/SONARP/ORACLE/data09/SONAR_CLOB_TS_38.dbf 20831551488
CLOB_TS /DB07/SONARP/ORACLE/data09/SONAR_CLOB_TS_39.dbf 15658369024
CLOB_TS /DB07/SONARP/ORACLE/data10/SONAR_CLOB_TS_42.dbf 15717359616
CLOB_TS /DB07/SONARP/ORACLE/data10/SONAR_CLOB_TS_43.dbf 15655370752
CLOB_TS /DB07/SONARP/ORACLE/data11/SONAR_CLOB_TS_44.dbf 10485276672
CLOB_TS /DB07/SONARP/ORACLE/data11/SONAR_CLOB_TS_54.dbf 1751007232
CLOB_TS /DB07/SONARP/ORACLE/data12/SONAR_CLOB_TS_45.dbf 20943249408
CLOB_TS /DB07/SONARP/ORACLE/data12/SONAR_CLOB_TS_46.dbf 10485235712
CLOB_TS /DB07/SONARP/ORACLE/data12/SONAR_CLOB_TS_47.dbf 15679864832
CLOB_TS /DB07/SONARP/ORACLE/data13/SONAR_CLOB_TS_48.dbf 15705817088
CLOB_TS /DB07/SONARP/ORACLE/data13/SONAR_CLOB_TS_49.dbf 15716311040
CLOB_TS /DB07/SONARP/ORACLE/data13/SONAR_CLOB_TS_50.dbf 17663680512
CLOB_TS /DB07/SONARP/ORACLE/data13/SONAR_CLOB_TS_51.dbf 14117568512
CLOB_TS /DB07/SONARP/ORACLE/data14/SONAR_CLOB_TS_52.dbf 10103119872
CLOB_TS /DB07/SONARP/ORACLE/data14/SONAR_CLOB_TS_53.dbf 9710149632
CLOB_TS /DB07/SONARP/ORACLE/idx02/SONAR_CLOB_TS_35.dbf 20775657472

hwm in GB 566.8251877

NAME GBYTES USED FREE PCT_USED LARGEST
CLOB_TS 621.1875 330.9694748 290.2180252 53.28012473 17.90049744
DRSYS 0.9765625 0.212081909 0.764480591 21.7171875 0.202857971
EDGAR_DATA_TS 4.8828125 3.153709412 1.729103088 64.58796875 1.338478088
EDGAR_IDX_TS 9.765625 2.799858093 6.965766907 28.67054688 3.430610657
EDGAR_STGNG_TS 7.8125 0.806442261 7.006057739 10.32246094 2.54069519
Q_DATA 0.1953125 0.102462769 0.092849731 52.4609375 0.050468445
RBS 23.4375 10.74221802 12.69528198 45.83346354 0.01953125
RBS_BIG 15.625 7.63E-06 15.62499237 4.88E-05 15.03905487
SONAR_IDX_TS 29.296875 27.2549057 2.041969299 93.03007813 0.493614197
SONAR_LG_IDX_TS 85.9375 81.10351563 4.833984375 94.375 4.150390625
SONAR_LG_TBL_TS 100.5859375 93.1640625 7.421875 92.62135922 4.052734375
SONAR_MD_IDX_TS 34.1796875 17.55273438 16.62695313 51.35428571 3.375
SONAR_MD_TBL_TS 16.6015625 9.614257813 6.987304688 57.91176471 2.206054688
SONAR_SM_IDX_TS 0.48828125 0.245483398 0.242797852 50.275 0.242797852
SONAR_SM_TBL_TS 2.44140625 1.432373047 1.009033203 58.67 0.951049805
SONAR_SM_TS 193.8476563 176.3231049 17.52455139 90.95962689 1.175796509
SONAR_STGNG_DAT_TS 4.8828125 0.26171875 4.62109375 5.36 4.620117188
SONAR_STGNG_IDX_TS 1.46484375 0.170898438 1.293945313 11.66666667 1.29296875
SYSTEM 0.29296875 0.259887695 0.033081055 88.70833333 0.030357361
TEMP 34.1796875 2.988327026 31.19136047 8.742991071 0.009765625
TEST 4 7.63E-06 3.999992371 0.000190735 3.999992371
TEXT_IDX_TS 118.1640625 114.0585861 4.105476379 96.52561338 0.1953125
USERS 0.48828125 0.151298523 0.336982727 30.9859375 0.313522339
USER_TOOLS 4.8828125 4.409286499 0.473526001 90.3021875 0.363670349
VSA 0.029296875 0.018554688 0.010742188 63.33333333 0.010742188

1315.646484 877.7952576 437.8512268 1261.695107 68.00608063

Hi Tom:

I tried your stored procedure show_space as well as from user_tables to generate the used/free/unused blocks of a table.

Here is what I got.

set serveroutput on
exec show_space('ITM_DIMENSION');

Free Blocks............................. 10,776
Total Blocks............................ 10,880
Total Bytes............................. 89,128,960
Total MBytes............................ 85
Unused Blocks........................... 103
Unused Bytes............................ 843,776
Last Used Ext FileId.................... 94
Last Used Ext BlockId................... 51,977
Last Used Block......................... 25
PL/SQL procedure successfully completed.

But, if I run

select blocks, empty_blocks from user_tables where table_name = 'ITM_DIMENSION';

BLOCKS EMPTY_BLOCKS
---------- ------------
10776 103

The same result comes back even after I run

exec dbms_stats.gather_table_stats(ownname=>'DBA', tabname=>'ITM_DIMENSION', estimate_percent=>100);

How come the stored procedure reports 10776 as the free blocks?

Then, I tried to create a second table

create table ITM_DIMENSION_TEST as select * from ITM_DIMENSION;

I got

exec show_space('ITM_DIMENSION_TEST');

Free Blocks............................. 0
Total Blocks............................ 5,376
Total Bytes............................. 44,040,192
Total MBytes............................ 42
Unused Blocks........................... 25
Unused Bytes............................ 204,800
Last Used Ext FileId.................... 131
Last Used Ext BlockId................... 1,301,641
Last Used Block......................... 103
PL/SQL procedure successfully completed.

exec dbms_stats.gather_table_stats(ownname=>'DBA', tabname=>'ITM_DIMENSION_TEST', estimate_percent=>100);

select blocks, empty_blocks from user_tables where table_name = 'ITM_DIMENSION_TEST';


BLOCKS EMPTY_BLOCKS
---------- ------------
5350 0


I could not interpret the results. Any ideas? Thanks.

ops$tkyte%ORA10GR2> create table t1
  2  tablespace manual
  3  as
  4  select *
  5    from all_objects
  6   order by object_id
  7  /

Table created.

ops$tkyte%ORA10GR2>
ops$tkyte%ORA10GR2> delete from t1
  2   where rowid in
  3  (select rid
  4     from (select rowid rid, row_number() over (order by rowid) rn
  5             from t1
  6                  )
  7   where mod(rn,3) in (0,1)
  8  );

34128 rows deleted.

ops$tkyte%ORA10GR2> commit;

Commit complete.

ops$tkyte%ORA10GR2>
ops$tkyte%ORA10GR2> create table t2
  2  tablespace manual
  3  as
  4  select * from t1;

Table created.

ops$tkyte%ORA10GR2>
ops$tkyte%ORA10GR2> exec show_space( 'T1' )
Free Blocks.............................             705
Total Blocks............................             768
Total Bytes.............................       6,291,456
Total MBytes............................               6
Unused Blocks...........................              62
Unused Bytes............................         507,904
Last Used Ext FileId....................              10
Last Used Ext BlockId...................               9
Last Used Block.........................              66

PL/SQL procedure successfully completed.

ops$tkyte%ORA10GR2> exec show_space( 'T2' )
Free Blocks.............................               0
Total Blocks............................             256
Total Bytes.............................       2,097,152
Total MBytes............................               2
Unused Blocks...........................              20
Unused Bytes............................         163,840
Last Used Ext FileId....................              10
Last Used Ext BlockId...................             137
Last Used Block.........................             108

PL/SQL procedure successfully completed.

Hi Tom.
I have a simple question. Could I use alter table T shrink to improve performance? Let´s suppose that my table has 5 million rows and 2 million rows were deleted. Is it possible that table shrink can improve performance in queries that use that table?
Thanks in advance.

Tom,

Specifically, I have a CLOB in a table that my client is slowly deleting the contents of for each record. They want to know how much space will be reclaimed at any given time.

In short, there is no way to tell how many bytes will be reclaimed under the HWM (v9.2.0.7) before the table is reorg'd (ie, copied/truncated/inserted or ALTER TABLE ... MOVE or in this case, just moving the LOB.)? I see the Freelist count but as you state, there is certainly no certainty that these blocks are even remotely empty.

Hi Tom,

We have a table size is 76MB and stable, but yesterday it grows to 4.3G. Please see the below show_space. Do you know how this table could blow upto 4.3GB by just deleting existing 188,965 rows and inserting 189,073 new rows? It was running like this everyday, but only yesterday it grow expectedly. After move the table it returns to 76MB. Thanks!

Unformatted Blocks ..................... 534,926
FS1 Blocks (0-25) ..................... 8,468
FS2 Blocks (25-50) ..................... 0
FS3 Blocks (50-75) ..................... 0
FS4 Blocks (75-100)..................... 8,591
Full Blocks ..................... 974
Total Blocks............................ 553,984
Total Bytes............................. 4,538,236,928
Total MBytes............................ 4,328
Unused Blocks........................... 0
Unused Bytes............................ 0
Last Used Ext FileId.................... 7
Last Used Ext BlockId................... 996,361
Last Used Block......................... 256

Hi Tom,

We are experiencing the same problem. According to bug 7299727 (see metalink) this is related to bug 5890312. Unfortunately, that last bug is not open to the public (yet).

Having said that, I would like to understand how an unformatted block below the HWM can be created. When I just delete records, de blocks are marked empty (FS4), not unformatted. Any ideas?

Example:

Connected to:
Oracle9i Enterprise Edition Release 9.2.0.8.0 - 64bit Production
With the Partitioning, OLAP and Oracle Data Mining options
JServer Release 9.2.0.8.0 - Production

SQL> create table arbor.lennert (id number) tablespace cat_ref;

Table created.

SQL> @show_space ARBOR LENNERT TABLE
Unformatted Blocks .....................0
FS1 Blocks (0-25)  .....................0
FS2 Blocks (25-50) .....................0
FS3 Blocks (50-75) .....................0
FS4 Blocks (75-100).....................0
Full Blocks        .....................0
Total Blocks............................16
Total Bytes.............................131072
Total MBytes  ..........................0
Unused Blocks...........................13
Unused Bytes............................106496
Last Used Ext FileId....................18
Last Used Ext BlockId...................321545
Last Used Block.........................3

PL/SQL procedure successfully completed.

SQL> insert into arbor.lennert (select rownum from dba_objects);

7180 rows created.

SQL> commit;

Commit complete.

SQL> @show_space ARBOR LENNERT TABLE
Unformatted Blocks .....................0
  FS1 Blocks (0-25)  .....................1   !
FS2 Blocks (25-50) .....................0
FS3 Blocks (50-75) .....................0
  FS4 Blocks (75-100).....................2   !
  Full Blocks        .....................10  !
Total Blocks............................16
Total Bytes.............................131072
Total MBytes  ..........................0
Unused Blocks...........................0
Unused Bytes............................0
Last Used Ext FileId....................18
Last Used Ext BlockId...................321545
Last Used Block.........................16

PL/SQL procedure successfully completed.

SQL> delete from arbor.lennert;

7180 rows deleted.

SQL> commit;

Commit complete.

SQL> @show_space ARBOR LENNERT TABLE
Unformatted Blocks .....................0
FS1 Blocks (0-25)  .....................0
FS2 Blocks (25-50) .....................0
FS3 Blocks (50-75) .....................0
  FS4 Blocks (75-100).....................13   !
Full Blocks        .....................0
Total Blocks............................16
Total Bytes.............................131072
Total MBytes  ..........................0
Unused Blocks...........................0
Unused Bytes............................0
Last Used Ext FileId....................18
Last Used Ext BlockId...................321545
Last Used Block.........................16

PL/SQL procedure successfully completed.

Thank you in advance!

Sir
Regarding shrinking the HWM,
Assume I have thousands of table,
How can I know which table needs reorganization?
How can I decide wheter I need to shrink the hwm of the tables?

You can use EM
or
Select * from table (dbms_space.asa_recommendations())

Tom ,

If I have 4 billion records in a table , in this table , around 100,000 records either updated or inserted daily

And if I perform delete operations of 100,000 records daily ,

I presume I will have a defragmentation .


By reading through the documentation , shrinking the segments will reset the HWM , colaseces the space ...
thus ful table scan need not go too far ( blocks containing no data) ..

Would it also help index scan or improve performance?

Adding to the above question .

We use ASM and we are in 10.2.0.4 on RHEL

Excellent explanation of HWM.
:)

hi Tom,

I have 10-15 tables that have 3-5 BLOB columns in it and has milions of record. Since the record size incresed very high so it was decided to delete the records that are 5 years old. Now deleting those records didnt reset the HWM for which lot of records were deleted.

Now Performace degraded because of that.. We want to reset the HWM but very long downtime is not allowed. How can we achive that.

Hi TOm,

thanks for the reply.. the requirement is such that we have to reset the HWM any how. So I was wondering if we could move the tablespace or deallocate unused...

yes the performance has degraded as around 80% rows were deleted..Please help

Thanks Soooo much for such a great site...Really Tom your site is a great help to all the new DBA's like me ;)
The way the HWM concept is explained with example is just too good....Thanks a ton :)

Hi Tom,

I have gone through the process to find the fragmented blocks which you have explained previously.
I have a doubt that a block may contain more than one row. Let it be 5 per block. If i
delete 4 rows from that table, then that block contains only 1 row.

The space released after deleting the 4 rows is fragmented which cannot be used for
further insertions. This block is under the used blocks(as it contains 1 row).

But actually this is a fragmented block. How can we get these type of fragmented blocks.
Please advice if i am going in a wrong way.

Thanks in advance,
Venkat

Hi Tom,

I have 1 table which get flushed and loaded daily. It has close to 1 million records. We want to replicate the changes into Datawarehouse so we have to use delete (not truncate).
How can we reset the HWM before we load the table?

Hi Tom,
Does the "gap" between the number of rows vs. HWM have any impact on sql plan. Example:
1. row count: 1mil, HWM=50,000
2. after delete, row count: 10,000, HWM=50,000 still the same
Will sql plan be different? When the table is involved in a full scan, HWM can have a significant effect on the execution time. So, does the optimizer do any evaluation of the number of blocks with data vs. HWM to determine the cost of the full scan?

Hi Tom,

I have a question regarding usage of HWM
I am trying to copy data from master table from remote DB to local machine first day there are 698 rows and the second day there are 709 rows can i copy the 9 new rows that have been yesterdays HWM to todays updated HWM directly or can the copy be done only based on the column on the master table which captures date when a row was updated