Binary search on SAP DB sorted tables

Use index table key in your Select statement ! – this is very clear and correct advise …unfortunately not always easy to achieve it !

Most often WM / SD / FI are the departaments which are in the top of transactions recorders . Their Docu Header Tables ( those which are ending in K – from german Kopf) is having the binom Docu Number – Date of creation available , and , in all cases Docu Number is index key ! Most of our sql queries will rely on a lot of different secondary key and for sure there will be a ‘Where date in [] ‘ .

My aim is to build a FM which must have as import parameters a table name , key field and date name field and export a min and max key field which can be used in a further selection . First I will test the logic on a single table using 2 routines . With help of your comments & advises I should be able to build the final version.

In example below I am trying to determine a minimum and a maximum document number range , based on a date range , which can be used in a Sql select directive in order to increase the performances .

In my scarce resources sandbox system , LTAK counts almost 2 mil records while LTAP around 10 mil .

Selecting from LTAK only based on LGNUM and BDATU is like watching a 3 legs dog walking ! Of course , LTAP is doing even worse (assuming that you won’t use a inner join with LTAK or FAE )!

Trivial condition for binary search is to have an ascending sorted table , it means that document number must be generated by NUMBER_GET_NEXT which implies an entry in NRIV table for this specific number range.

LTAK is matching the case .

Binary search approach :

According to picture bellow we have to find all TANUM which are laying in a given date range ( s_datum ) in Selection-Screen and for a known LGORT .

My approach is to find a TANUM at left of s_datum-low and one at right of s_datum-high . Of course , first TANUM and last TANUM in LTAK table would fulfill it , but won’t increase select performances , therefore , we have defined a delta (type i) and build 2 ranges based on it. Intersection between any of delta_date range and s_datum range should be the null set . The length (delta) of delta_date range must be defined in such a way to avoid a large number of iterations in order to find an entry inside of it . It is easy observable that the length and number of iterations are inversely proportional . A small delta would find a TANUM closer to one of the s_datum ‘s edges but the number of hits would increase , viceversa , a big delta would return in less iterations a TANUM which may not be too close to s_datum’s edges .

Coding :

1 . Defining variabiles and selection-screen :

in structures gs_min_ltak and gs_max_ltak we are recording first and last entry in LTAK table ;

min_tanum and max_tanum structures will hold the found values for which we are looking for ;

delta_date is the range in which we will search for min_tanum and max_tanum ;

delta (type INT1/{0} ) is the legth of delta_date interval .

2 . Routine find_min_max_tanum_in_ltak is responsabile for binary search recursivity .

1^st  step would be to read extremities of table ( min and max tanum )

  SELECT SINGLE * FROM ltak INTO gs_min_ltak WHERE lgnum = p_lgnum.
SELECT  * FROM ltak UP TO 1 ROWS INTO gs_max_ltak WHERE lgnum = p_lgnum ORDER BY tanum DESCENDING.
ENDSELECT.

2^nd  , based on length of interval , we would determine a maximum number of failed iterations . ( generously i choose a max of 2 x (log base 2 of N )    where N is number of rec in LTAK ) .

max_iterations =  2 * log( gs_max_ltak–tanum – gs_min_ltak–tanum ) / log( 2 ) .

If number of failed iterations would be greater than an expected number of iterations , then we might consider the existence of noise in data distribution ( mass documents deletion , long period of time without records ..) or our delta value was setted up in such a way to find posted records during the weekend ( which may have a low value of probability density function for this interval).

3^rd set_delta_date_range routine is responsabile for setting the boundaries of left and right intervals,where we have to find our records .

4^th find_tanum routine is the recursive binary search routine .

Exemple of SAT measurement

I have measured runtime of finding 1 record in LTAK table in a given period . It took almost 12 secondes .

Same interogation was done in binary search in my Z report . It took 0,1 seconds to find 2 TANUMs (left and right of interval ) . Please see bellow results .

SAT – SE16N / LTAK Table

SAT – SE16N Measurement

SAT – Z_MIN_MAX_LTAK

SAT – Z_MIN_MAX_LTAK measurement

Z_MIN_MAX_LTAK output

*&———————————————————————*
*& Report  Z_MIN_MAX_LTAK
*&
*&———————————————————————*
*&
*&
*&———————————————————————*

report z_min_max_ltak.

data : gs_min_ltak type ltak ,
gs_max_ltak type ltak.
data : min_tanum type ltak ,
max_tanum type ltak .
data : number_iterations type i,
target type string ,
lv_error type xflag.
data : delta_date type range of sy–datum with header line .

data max_iterations type i.
select-options : s_datum for sy–datum obligatory.
parameters : p_lgnum type lgnum default ‘SVT’ obligatory,
p_delta type int1 default 8 obligatory. ” In my case 8 days are the maximum which can exists without records .

at selection-screen .
if s_datum–high is  initial.
message e031(/iwbep/epm_product) with ‘Date to ‘.
*   &1 is mandatory
endif.
if p_delta = 0.
message e094(axt_model) with ‘Delta’.
*   Field &1: Enter a length greater than 0
endif.
start-of-selection .
perform find_min_max_tanum_in_ltak .
perform write.

*&———————————————————————*
*&      Form  FIND_MIN_MAX_TANUM_IN_LTAK
*&———————————————————————*
*       text
*———————————————————————-*
*  –>  p1        text
*  <–  p2        text
*———————————————————————-*
form find_min_max_tanum_in_ltak .

select single * from ltak into gs_min_ltak where lgnum = p_lgnum.
select  * from ltak up to 1 rows into gs_max_ltak where lgnum = p_lgnum order by tanum descending.
endselect.
max_iterations =  2 * log( gs_max_ltak–tanum – gs_min_ltak–tanum ) / log( 2 ) .
if s_datum–low < gs_min_ltak–bdatu and s_datum–high < gs_min_ltak–bdatu.
message e048(ei).
*   No data selected for this time period
elseif s_datum–low > gs_max_ltak–bdatu  and s_datum–high > gs_max_ltak–bdatu  .
message e048(ei).
*   No data selected for this time period
endif.

if s_datum–low <= gs_min_ltak–bdatu .
min_tanum = gs_min_ltak.
else.
target = ‘MIN’ .
perform set_delta_date_range using target.
perform find_tanum using gs_min_ltak–tanum gs_max_ltak–tanum target.
endif.

if s_datum–high >= gs_max_ltak–bdatu.
max_tanum = gs_max_ltak.
else.
target = ‘MAX’.
perform set_delta_date_range using target.
perform find_tanum using min_tanum–tanum gs_max_ltak–tanum target.
endif.
if lv_error is not initial.
message ‘ Too many iterations .Increase delta !’ type ‘E’ display like ‘E’ .
endif.
endform.                    ” FIND_MIN_MAX_TANUM_IN_LTAK
*&———————————————————————*
*&      Form  FIND_TANUM
*&———————————————————————*
*       text
*———————————————————————-*
*      –>P_MIN  text
*      –>P_MAX  text
*      –>P_TARGET  text
*———————————————————————-*
form find_tanum  using  p_min type tanum
p_max type tanum
p_target type string.

data :temp_tanum type tanum ,
temp_ltak type ltak.
number_iterations = number_iterations + 1 .
temp_tanum = ( p_max – p_min ) / 2 + p_min .
select single * from ltak into temp_ltak where lgnum = p_lgnum and tanum = temp_tanum .
if sy–subrc <> 0.
select single * from ltak into temp_ltak where lgnum = p_lgnum and tanum > temp_tanum .
case p_target.
when ‘MIN’.
perform find_tanum using temp_ltak–tanum p_max p_target.
when ‘MAX’.
perform find_tanum using p_min temp_ltak–tanum p_target.
when others.
endcase.
else.
***********delta too small or too many gaps ( records are not uniform distributed)
if number_iterations > max_iterations . ” Empty interval
case target.
when ‘MIN’.
min_tanum–tanum = p_min.
lv_error = ‘X’.
exit.
when ‘MAX’.
max_tanum–tanum = p_max .
lv_error = ‘X’.
exit.
when others.
endcase.
endif.
**********************************
if temp_ltak–bdatu in delta_date.
case target.
when ‘MIN’.
min_tanum = temp_ltak.
exit.
when ‘MAX’.
max_tanum = temp_ltak.
exit.
when others.
endcase.
elseif temp_ltak–bdatu < delta_date–low.
perform find_tanum using temp_ltak–tanum p_max target.
elseif temp_ltak–bdatu > delta_date–high .
perform find_tanum using p_min temp_ltak–tanum target.
endif.
endif.
endform.                    ” FIND_TANUM
*&———————————————————————*
*&      Form  SET_DELTA_DATE_RANGE
*&———————————————————————*
*       text
*———————————————————————-*
*      –>P_TARGET  text
*———————————————————————-*
form set_delta_date_range  using p_target type string.
free delta_date.
case p_target.
when ‘MIN’.
delta_date–low = s_datum–low – p_delta .
delta_date–high = s_datum–low – 1.
delta_date–sign = ‘I’.
delta_date–option = ‘BT’.
append delta_date.
when ‘MAX’.
delta_date–low = s_datum–high + 1.
delta_date–high = s_datum–high + p_delta .
delta_date–sign = ‘I’.
delta_date–option = ‘BT’.
append delta_date.
when others.
endcase.
endform.                    ” SET_DELTA_DATE_RANGE
*&———————————————————————*
*&      Form  WRITE
*&———————————————————————*
*       text
*———————————————————————-*
*  –>  p1        text
*  <–  p2        text
*———————————————————————-*
form write .
write :/ ‘Our goal was to find a min TANUM and a max TANUM in LTAK table ,’.
write :/ ‘positioned to the left of s_datum-low = ‘ ,s_datum–low , ‘ and to the right side of s_datum-high = ‘ ,s_datum–high , ‘ within a range (delta) of max ‘, p_delta centered, ‘days.’ .
skip.
write :/ ‘Goal was achieved in a # of iterations = ‘ ,number_iterations no-zero left-justified , ‘Number of max allowed iterations = ‘ , max_iterations.
write :/ ‘Min TANUM= ‘ , min_tanum–tanum no-zero left-justified .
write :/ ‘Min TANUM date = ‘ , min_tanum–bdatu left-justified.
write :/ ‘Max TANUM= ‘ , max_tanum–tanum no-zero left-justified.
write :/ ‘Max TANUM date ‘ , max_tanum–bdatu left-justified.

endform.                    ” WRITE