For All Entries is NOT better than INNER JOIN in most cases

Question

I quote from Siegfried Boes' excellent post here: Will writing an inner join be better or creating a view?
For all the FOR ALL ENTRIES lovers ... there is no proof for these reappearing recommendation.
There is nearly nobody who receives forum points, who recommends FOR ALL ENTRIES instead of Joins. What is the reason ???
It is easier to prove the opposite. A Join is a nested loop inside the database, a FOR ALL ENTRIES is partly outside of the database. FOR ALL ENTRIES works in blocks, joins on totals.
FOR ALL ENTRIES are not recommded on really large tables, because the chances are too high that
too many records are transferred.
People prefer FOR ALL ENTRIES, because JOINs are not so easy to understand. Joins can go wrong, but with a bit of understanding they can be fixed.
Some Joins are slow and can not be fixed, but then the FOR ALL ENTRIES would be extremely slow.
There are several kinds of views:
- projection views, i.e. only one table involved just fields reduced
- join views, several tables, joins conditions stored in dictionary
- materialized views, here the joined data are actually stored in the database. Storing and synchronisation has to be done manually.
Only the last one creates real overhead. It should be the exception.
Join Views and Joins are nearly identical. The view is better for reuse. The join is better in complicated, becuase if the access goes wrong, it can often be fixed by adding a hint. Hints can not be added to views.
Abraham Bukit  points out:
If it is cluster table, (you can't use join). If it is buffered table, I would also say avoid join.
If they all are transaction table which are not buffered and are not cluster tables.  
He further supports Siegfried's statement that FAE is easier to undestand than INNER JOINs.
Thomas Zloch says, regarding buffered tables:
At least think twice, maybe compare runtimes if in doubt. 
So, unless someone has some EVIDENCE that FOR ALL ENTRIES is better, I don't think we want to see this discussed further.
Kind regards
Matt

Answer 1

To give food for thought here's an example I  gave in a thread:
If you have a statement like
SELECT ... FOR ALL ENTRIES IN FAE_itab WHERE f = FAE_itab-f.
SAP sends it to the database depending how the parameter rsdb/prefer_union_all is set:
rsdb/prefer_union_all = 0 =>
SELECT ... WHERE f = FAE_itab[1]-f
          OR    f = FAE_itab[2]-f
          OR    f = FAE_itab[N]-f
You have some influence  of the generated statement type: Instead of OR'ed fields an IN list can be used
if you have only a single coulmn N to compare:
rsdb/prefer_in_itab_opt parameter:
SELECT ... WHERE f IN (itab[1]-f, itab[2]-f, ..., itab[N]-f)
rsdb/prefer_union_all = 1 =>
SELECT ... WHERE f = FAE_itab[1]-f
UNION ALL SELECT ... WHERE f = FAE_itab[2]-f
UNION ALL SELECT ... WHERE f = FAE_itab[N]-f
see: Note 48230 - Parameters for the SELECT ... FOR ALL ENTRIES statement
As you can see for the 2nd parameter several statements are generated and combined with a UNION ALL,
the first setting generates statements with OR's (or uses IN  if possible) for the entries in FAE_itab.
I give you a little example here (my parameters are set in a way that the OR's are translated to IN lists; i traced the execution in ST05)
Select myid into table t_tabcount from mydbtable
  for all entries in t_table    " 484 entries
    where myid = t_table-myid .
ST05 trace:
|Transaction SEU_INT|Work process no 0|Proc.type  DIA|Client  200|User |
|Duration |Obj. name |Op.    |Recs.|RC    |Statement|
| 640|mydbtable |PREPARE|   |  0|SELECT WHERE "myid" IN ( :A0 , :A1 , :A2 , :A3 , :A4 ) AND "myid" = :A5|
| 2|mydbtable |OPEN   |   |  0|SELECT WHERE "myid" IN ( 1 , 2 , 3 , 4 , 5 ) AND "myid" = 72 |
| 2.536|mydbtable |FETCH  |    0|  1403|   |
| 3|mydbtable |REOPEN |   |  0|SELECT WHERE "myid" IN ( 6 , 7 , 8 , 9 , 10 ) AND "myid" = 72 |
| 118|mydbtable |FETCH  |  0|  |
| 2|mydbtable |REOPEN |  |  0|SELECT WHERE "myid" IN ( 11 , 12 , 13 , 14 , 15 ) AND "myid" = 72     |
| 3|mydbtable |REOPEN |  |  0|SELECT WHERE "myid" IN ( 475 , 476 , 477 , 478 , 479 ) AND "myid" = 72  |
| 94|mydbtable |FETCH  | 0| 1403|   |
| 2|mydbtable |REOPEN |   |  0|SELECT WHERE "myid" IN ( 480 , 481 , 482 , 483 , 484 ) AND "myid" = 72 |
You see the IN list contained 5 entries each , wich made up about 97 statements for all 484 entries.
For every statment you have a single fetch operation wich means a separate access to the database.
If you would replace the FAE with a join you would only have one fetch to the database.
With the example above we can derive these observations:
1. From database point of view these settings kill performance when you access a big table and/or have a lot of entries or columns in your FAE_itab. Furthermore, you hide information what data you will access
at all and thus you block the database from creating a more efficient execution plan because it DOESN'T KNOW wich data you will select in the next step. I.e. it may be more efficient to scan the table in one shot instead of having many index accesses - but the database can make this decision only if it can examine ONE statement that has ALL the information of what data to retrieve.
2. A second impact is that with every statement execution you trigger the allocation of database resources
wich will contribute to the overhead described above.
Said that, FAE  can never be a replacement for joining big tables (think of having a table with thousands of records in a FAE table )
Edited by: kishan P on Nov 2, 2010 2:16 PM - Format Fixed