Exadata sizing updated for 3tb drives 1/2 rack SATA

OK, Now I new Exadata coming in that has 3tb drives, and the first question asked is .. How much disk to I have to configure on it ?  Well I'm going to expand on a previous entry I did on sizing .

1/2 Rack. Sata drives. normal redundancy

This means we have

• 7 storage cells
• Each storage cell contains 12 disks
• each disk is 3tb (which is about 2.794 tb usable)  *** This is calculated using base 1024 
• The first 2 disks in each storage cell has 29.103g already partitioned for the OS (which is mirrored).
• The rest of the disks in the group are used for DBFS

Given this, I am going to calculate out the total disk available then subtract out the 29.103g (for OS and DBFS).

First 12 disks * 7 cells x 2.794 = 234.696 tb of total raw storage/
Subtract out 29g* 2 disks * 7 cells = 406g    ----- OS
Subtract out 29g * 10 disks * 7 cells = 2.03tb  -----   DBFS
Available raw is 234.696 - 2.436 = 232.26  

Now I said we were running Normal Redundancy.. This means that we loose 1/2

DBFS = 1.015tb
OS        29g
Remaining for Data and Reco = 116.13

But of course we need to account for cell being off line.  This takes out 1/7 of the storage.

DBFS   === .870 tb (29g * 10 * 6)/2
Everything else  ===  ( 2.765 * 12 disks * 6 cells)/2   == 99.54

So now we have 99.54 raw storage available for Data and Reco.

This is now easy to figure out now.. You have really 100tb raw storage (with normal redundancy) to split up between Data and Reco.

Now a full rack is easy to do.

2.765 * 12 disks * 13 cells) / 2 =  215.67tb 

What extended stats do I have on my database?

I've been starting to work with Extended statistics to help the optimizer find the best plan. This is a great feature that is outlined by @sqlmaria (Maria Coogan) here.

But once you create extended statistics, how do you know what is there ?  I wrote this query to find out what function based indexes, what extended statistics, and what their definition are. 

Here is my script.

column table_owner alias "owner" format a15
column table_name alias  "Table Name" format  a30
column function_index alias  "F Index" format  a8
column Index_name  alias  "Index Name"  format a30
column data_default alias  "Definition"  format a50
set pagesize 1000
select table_owner,
         table_name,
        nvl2(index_name,'YES','NO') function_index,
        index_name,
        data_default
        from
        (
select owner table_owner,table_name,
(select distinct index_name from dba_ind_columns b where a.column_name=b.column_name and a.owner=b.index_owner and a.table_name=b.table_name) index_name
,data_default
-- ,     DBMS_LOB.SUBSTR( to_lob(data_default),100,1)
 from dba_tab_cols a
  where virtual_column='YES' and hidden_column='YES'  and (owner not in ('SYS','WMSYS','XDB','SYSMAN','MDSYS','EXFSYS','PR_MDS') and owner not like 'APEX_%')
  )
order by table_owner,table_name;

and this is what the output looks like..

TABLE_OWNER     TABLE_NAME                     FUNCTION INDEX_NAME                     DATA_DEFINITION
--------------- ------------------------------ -------- ------------------------------ --------------------------------------------------
BGRENN          TAB_SCHR_PERD                      NO                                  COALESCE("COL1","COL2")
BGRENN          TAB2                              YES   IDX_TAB2                       "COL1"||' '||"COL2"
BGRENN          TAB3                               NO                                  COALESCE("COL1","COL2")
BGRENN          TAB4                              YES   IDX_TAB4                       COALESCE("COL1","COL2",0)
BGRENN          TAB4                              YES   IDX_TAB4                       COALESCE("COL3",0)
BGRENN          TAB5                              YES   IDX_TAB5                       COALESCE("COL1","COL2",0)
BGRENN          TAB6                              YES   IDX_TAB6                       NVL("COL1",(-1))
BGRENN          TAB6                              YES   IDX_TAB6                       NVL("COL2",(-1))
BGRENN          TAB6                              YES   IDX_TAB6                       NVL("COL3",(-1))
BGRENN          TAB6                              YES   IDX_TAB6                       NVL("COL4",'x')
BGRENN          TAB7                              YES   IDX_COMPOSITE                  "COL1"
BGRENN          TAB7                              YES   IDX_COMPOSITE                  "COL3"

Notice the Function colunmn. This is a "YES" or "NO" depending on if this is a function based index, or just extended statistics.

This should help tell where your extended statistics are in your database.

Exadata tips

I wanted to write up an Exadata tip that I learned.

Background :  I wanted to do a simple "select count(1) from mytable".  mytable has a primary key on it.  The count seemed to be taking a long time for an Exadata.

First the "select count(1) from mytable". You can see that it uses an index storage fast full scan.  The top wait event is "cell multiblock physical read".  The query does  2 Million Disk reads in 3 Minutes 28 seconds.

But this seems slow...

select count(1)
from
     MYTABLE MYTABLE

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        2     56.46     208.27    2030378    2031797         20           1
------- ------  -------- ---------- ---------- ---------- ----------  ----------
total        4     56.46     208.27    2030378    2031797         20           1

Misses in library cache during parse: 1
Optimizer mode: ALL_ROWS
Parsing user id: SYS
Number of plan statistics captured: 1

Rows (1st) Rows (avg) Rows (max)  Row Source Operation
---------- ---------- ----------  ---------------------------------------------------
         1          1          1  SORT AGGREGATE (cr=2031797 pr=2030378 pw=0 time=208276577 us)
 592848893  592848893  592848893   INDEX STORAGE FAST FULL SCAN PK_MYTABLE (cr=2031797 pr=2030378 pw=0 time=245927651 us cost=523897 size=0 card=572441788)(object id 312310)


Elapsed times include waiting on following events:
  Event waited on                             Times   Max. Wait  Total Waited
  ----------------------------------------   Waited  ----------  ------------
  library cache lock                              4        0.00          0.00
  Disk file operations I/O                       42        0.00          0.00
  library cache pin                               2        0.00          0.00
  SQL*Net message to client                       2        0.00          0.00
  cell single block physical read                15        0.00          0.00
  cell list of blocks physical read               2        0.00          0.00
  cell multiblock physical read               15959        0.26        152.20
  latch: object queue header operation            1        0.00          0.00
  SQL*Net message from client                     2       10.38         10.38
********************************************************************************

Next I did a FTS.

"select /*+full(MYTABLE) */ count(1) from  MYTABLE MYTABLE ;

You can see this did 2.2 Million disk reads (more than the index scan), but the wait event is sql_net. With the "cell smart table scan", there were very few waits, and the wait was much shorter.

select /*+full(MYTABLE) */ count(1)
from
     MYTABLE MYTABLE  

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        2     65.11      66.48    2224642    2225028         21           1
------- ------  -------- ---------- ---------- ---------- ----------  ----------
total        4     65.11      66.48    2224642    2225028         21           1

Misses in library cache during parse: 1
Optimizer mode: ALL_ROWS
Parsing user id: SYS
Number of plan statistics captured: 1

Rows (1st) Rows (avg) Rows (max)  Row Source Operation
---------- ---------- ----------  ---------------------------------------------------
         1          1          1  SORT AGGREGATE (cr=2225028 pr=2224642 pw=0 time=66486729 us)
 592848893  592848893  592848893   PARTITION RANGE ALL PARTITION: 1 33 (cr=2225028 pr=2224642 pw=0 time=140325566 us cost=533066 size=0 card=572441788)
 592848893  592848893  592848893    TABLE ACCESS STORAGE FULL MYTABLE PARTITION: 1 33 (cr=2225028 pr=2224642 pw=0 time=54479242 us cost=533066 size=0 card=572441788)


Elapsed times include waiting on following events:
  Event waited on                             Times   Max. Wait  Total Waited
  ----------------------------------------   Waited  ----------  ------------
  SQL*Net message to client                       2        0.00          0.00
  Disk file operations I/O                      129        0.00          0.00
  gc current block 2-way                         65        0.00          0.00
  enq: KO - fast object checkpoint              101        0.00          0.02
  reliable message                               33        0.09          0.30
  gc current block 3-way                         13        0.00          0.00
  cell smart table scan                        1403        0.03          1.01
  gc cr block 3-way                               1        0.00          0.00
  gc current grant busy                          18        0.00          0.00
  gc cr block 2-way                              17        0.00          0.00
  gc cr multi block request                       5        0.00          0.00
  cell single block physical read                11        0.00          0.00
  cell list of blocks physical read               2        0.00          0.00
  gc cr grant 2-way                               3        0.00          0.00
  SQL*Net message from client                     2        9.62          9.62

Bottom line, if you want to a count on a table use the "FULL" hint.  The exadata is built for table scans, and this example shows that.

It also should make you rethink when to use indexes for an application, you see they can hurt you in some cases.