excessive increase in table sizing

[mariopellegrini]

I have a table sizing problem that I don't understand. doing a simple left join with a small table i added a variable to a table. The strange thing is that this new table with an extra variable (13 instead of 12) has gone from 8.54Gb to 21.17Gb! I can't figure it out, I think the physical memory should have increased only a little.

I report the complete results of the dictionary of the 2 tables in question (note the "filesize" variable)

[Reeza]

What is the length of the new character variable added?

[mariopellegrini]

200

[Reeza · Accepted solution]

Then yes, that is correct. 

You have 67,298,471 observations X 200 characters/bytes = 13,459,694,200 bytes = 13.5 GB

[ballardw]

Show the code for your "simple join".  It is likely that you added records, at a guess about doubling the number of records, because of the particular join.

Consider this small example. A "simple join" of two data sets of 6 observations each but the result has 16 observations.

data one;
  input a b;
datalines;
1  1
1  2
1  3
2  11
2  22
3  33
;

data two;
  input a z;
datalines;
1 111
1 222
1 333
2 1111
2 2222
2 3333
;

proc sql;
    create table example as
    select one.a, one.b, two.z
    from one 
         left join
         two 
         on one.a = two.a
   ;
quit;

Table size for many discussions is more about the number of observations and variables than bytes.

[mariopellegrini]

this is the code. I also report the data from the dictionary columns of the 2 tables in question

proc sql; 
create table RUOLO_DANNI2  as
select RU.* 
      ,C.DES_EDWH as RUOLO_DES
from RUOLO_DANNI1 RU
  left join  EDWH_ODS.CODICE_TIPO_RUOLO C
     on 
             RU.COD_TIPO_RUOLO = C.COD_EDWH 
			 and C.ts_inizio_validita  <= ru.ts_inizio_validita <=  C.ts_fine_validita
             and compress(C.cod_source)='1';
quit;

[Kurt_Bremser]

You added 200 bytes to your obslen, indicating that the new variable is $200 in length. This usually happens when the new variable is assigned a value from a function without properly defining it beforehand with a LENGTH statement.

Please show your code.

[mariopellegrini]

I simply brought the variable forward from another table without specifying the length, thus inheriting the original length

proc sql; 
create table RUOLO_DANNI2  as
select RU.* 
      ,C.DES_EDWH as RUOLO_DES
from RUOLO_DANNI1 RU
  left join  EDWH_ODS.CODICE_TIPO_RUOLO C
     on 
             RU.COD_TIPO_RUOLO = C.COD_EDWH 
			 and C.ts_inizio_validita  <= ru.ts_inizio_validita <=  C.ts_fine_validita
             and compress(C.cod_source)='1';
quit;

[Reeza]

,C.DES_EDWH as RUOLO_DES LENGTH=$20

Or whatever makes sense for you to try and reduce the size.

[Kurt_Bremser]

Well, an almost tripling of the observation size will also result in tripling the dataset size.

Use the COMPRESS=YES dataset option if the new variable is mostly empty.

[Tom]

Use the COMPRESS=YES system option (or COMPRESS=YES dataset option if you only want it to apply to one dataset) and it will take a lot less space on the disk.

Why attach a spreadsheet to share 8 numbers?

memname	nobs	obslen	nvar	compress	filesize
RUOLO_DANNI1	67,298,471	136	12	NO	9,169,469,440
RUOLO_DANNI2	67,298,471	336	13	NO	22,734,503,936