turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- SAS Programming
- /
- Base SAS Programming
- /
- Join Many Tables to One Master Table

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

11-15-2016 10:18 PM

I have a base master table of about 150 million rows. I have a series of about 20 different tables that I need to link into the master table in order to fill out additional columns. These additional tables range in size from 500 rows to 30 million rows. Each one has a unique key that can be used to link to the master table. What is my best bet in creating one massive base table with all the columns I need filled out? I have tried a series of inner and left joins through proc sql but it takes way too long, probably 20-30 hours if I let it continue to run.

I am a SAS novice but have a good understanding of SQL. Any tips for me?

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

11-15-2016 10:53 PM

Try to make use of user formats that will help you to avoid joining some of the tables. There are ways to use proc format for data merges. This solution would not be applied for all 20 tables but atleast you reduce some joins.

Suppose I have column having city name and I want to join to table to get country name. This can be done using proc format and you can avoid merging of table.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

11-16-2016 12:49 AM

If you have lot of RAM try data step hash tabled. Should be fast, but a bit more complicated programming compared to SQL.

Data never sleeps

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

11-16-2016 12:53 AM

I concur with @RahulG It sounds like you have a fact table and corresponding dimensions. If any are slowly changing dims you'll need to join those but for the others you should be able to apply format and avoid a join.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

11-16-2016 04:02 AM - edited 11-16-2016 04:03 AM

If you really must do this in SQL, and there is a common id var, then you might save time by joining the smaller tables first. then join the result to the master table.

Are all the tables to be joined on the same ID variable? And is it unique?

Another alternative is to consider a DATA step with a merge statement. Again if (1) you have a common ID var accross all tables, and (2) ID's are unique within each table, then you could do the following;

1. Sort each dataset by ID, then run this data step:

data want;

merge master table1 table2 .... table20;

by id;

run;

The issue will be how long sorting those data sets will take..

regards,

Mark

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

11-16-2016 08:36 AM

You want to "Create one massive base table with all the columns I need filled out." Do you want to insert Rows from the 20 different tables into the Master table based on the Unique Key? Are these Rows from 20 tables FILL the EMPTY slots(assuming that Unique Keys are spreadout in the Master)? Will there be any collision possible? Does your unique is Numeric?

Can you show some dummy records of Master? Also two of the other tables each not having more than 2 records.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

11-16-2016 10:26 AM

As you said your master table has 150 million and the tables to join has less than 30 million. I suggest you to use data step hash tables. This is the more efficient way. The good thing in using hash tables is the data is not needed to sort which saves lot of time.

Thanks,

Suryakiran

Suryakiran

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

11-17-2016 12:15 PM

```
SAS/WPS: Update master using 23 dimension tables 14 minutes not 20-30 hrs
This is not a big data problem.
inspired by
https://goo.gl/tzSYvK
https://communities.sas.com/t5/Base-SAS-Programming/Join-Many-Tables-to-One-Master-Table/m-p/311889
I have a base master table of about 150 million rows. I have a series of about 20 different
tables that I need to link into the master table in order to fill out additional columns.
These additional tables range in size from 500 rows to 30 million rows. Each one has a unique
key that can be used to link to the master table. What is my best bet in creating one massive base table
with all the columns I need filled out? I have tried a series of inner and left joins through
proc sql but it takes way too long, probably 20-30 hours if I let it continue to run.
I don't see how this can take 20-30hrs. It took 14 minutes on my $600 2008 Dell T7400,
probably cheaper now.
LOG
NOTE: The data set WORK.MASTER has 150,000,000 observations and 35 variables.
NOTE: DATA statement used (Total process time):
real time 14:18.11
cpu time 19:15.79
Adding descriptions from 23 dimension tables to one master table.
For small tables like these I do nto parallelize.
HAVE a MASTER table with 150 million obs
=========================================
NOTE: There were 150000000 observations read from the data set SPDE.MM150.
AND 23 Dimension tables
=======================
NOTE: There were 500 observations read from the data set SPDE.DIM500.
NOTE: There were 1000 observations read from the data set SPDE.DIM1000.
NOTE: There were 2000 observations read from the data set SPDE.DIM2000.
NOTE: There were 4000 observations read from the data set SPDE.DIM4000.
NOTE: There were 8000 observations read from the data set SPDE.DIM8000.
NOTE: There were 10000 observations read from the data set SPDE.DIM10000.
NOTE: There were 20000 observations read from the data set SPDE.DIM20000.
NOTE: There were 40000 observations read from the data set SPDE.DIM40000.
NOTE: There were 80000 observations read from the data set SPDE.DIM80000.
NOTE: There were 100000 observations read from the data set SPDE.DIM100000.
NOTE: There were 200000 observations read from the data set SPDE.DIM200000.
NOTE: There were 400000 observations read from the data set SPDE.DIM400000.
NOTE: There were 802140 observations read from the data set SPDE.DIM800000.
NOTE: There were 1000000 observations read from the data set SPDE.DIM1000000.
NOTE: There were 2000000 observations read from the data set SPDE.DIM2000000.
NOTE: There were 4054055 observations read from the data set SPDE.DIM4000000.
NOTE: There were 8333334 observations read from the data set SPDE.DIM8000000.
NOTE: There were 10000000 observations read from the data set SPDE.DIM10000000.
NOTE: There were 21428572 observations read from the data set SPDE.DIM20000000.
NOTE: There were 30000000 observations read from the data set SPDE.DIM30000000.
NOTE: There were 10000000 observations read from the data set SPDE.DIM10000000.
NOTE: There were 21428572 observations read from the data set SPDE.DIM20000000.
NOTE: There were 30000000 observations read from the data set SPDE.DIM30000000.
MASTER
Middle Observation(75000000 ) of SPDE.MM150 - Total Obs 150,000,000
14 variable, note the use of codes instead of descriptions
this is goo practice
-- CHARACTER --
C1 C 1 A
C2 C 1 A
C3 C 1 A
C4 C 1 A
C5 C 1 A
C6 C 1 A
C7 C 1 A
C8 C 1 A
C9 C 1 A
C10 C 1 A
-- NUMERIC --
N1 N 8 1111
N2 N 8 2222
N3 N 8 3333
N4 N 8 4444
KEY N 8 75000000
ONE OF THE DIMENSION TABLES
Up to 40 obs from SPDE.DIM500 total obs=500
Obs KEY DIM300000
1 1 MALE
2 300001 MALE
3 600001 MALE
4 900001 MALE
5 1200001 MALE
WANT NEW MATER TABLE with all the dimension information
Middle Observation(1 ) of Last dataset = WORK.XUNDER1MM - Total Obs 1
-- CHARACTER --
C1 C 1 A C1
C2 C 1 A C2
C3 C 1 A C3
C4 C 1 A C4
C5 C 1 A C5
C6 C 1 A C6
C7 C 1 A C7
C8 C 1 A C8
C9 C 1 A C9
C10 C 1 A C10
DIM300000 C 4 MALE DIM300000
DIM150000 C 2 VT DIM150000
DIM75000 C 3 OLD DIM75000
DIM37500 C 4 BLUE DIM37500
DIM18750 C 3 TOP DIM18750
DIM15000 C 3 LOW DIM15000
DIM7500 C 4 HIGH DIM7500
DIM3750 C 4 PRES DIM3750
DIM1875 C 4 BEST DIM1875
DIM1500 C 4 GROW DIM1500
DIM750 C 4 POOR DIM750
DIM375 C 4 RICH DIM375
DIM187 C 3 KID DIM187
DIM150 C 4 POOL DIM150
DIM75 C 4 MEAN DIM75
DIM37 C 3 MIN DIM37
DIM18 C 3 MAX DIM18
DIM15 C 3 OUT DIM15
DIM7 C 3 DEM DIM7
DIM5 C 3 GOP DIM5
-- NUMERIC --
KEY N 8 1 KEY
N1 N 8 1111 N1
N2 N 8 2222 N2
N3 N 8 3333 N3
N4 N 8 4444 N4
Join Many Tables to One Master Table
I have a base master table of about 150 million rows. I have a series of about 20 different
tables that I need to link into the master table in order to fill out additional columns.
These additional tables range in size from 500 rows to 30 million rows. Each one has a unique
key that can be used to link to the master table. What is my best bet in creating one massive base table
with all the columns I need filled out? I have tried a series of inner and left joins through
proc sql but it takes way too long, probably 20-30 hours if I let it continue to run.
I am a SAS novice but have a good understanding of SQL. Any tips for me?
%let pgm=utl_complex_update;
libname spde spde
('c:\wrk\spde_c','d:\wrk\spde_d','e:\wrk\spde_e','g:\wrk\spde_g','h:\wrk\spde_h')
metapath =('c:\wrk\spde_c\metadata')
indexpath=(
'c:\wrk\spde_c'
,'d:\wrk\spde_d'
,'e:\wrk\spde_e'
,'g:\wrk\spde_g'
,'h:\wrk\spde_h')
datapath =(
'c:\wrk\spde_c'
,'d:\wrk\spde_d'
,'e:\wrk\spde_e'
,'g:\wrk\spde_g'
,'h:\wrk\spde_h')
partsize=500m
;
* CREATE SOME DATA;
* it is not unusual to use codes in a fact table instead of long descriptions;
data spde.mm150(index=(key/unique));
array nums[4] n1-n4 (1111,2222,3333,4444);
array codes[10] $1 c1-c10 (10*'A');
do key=1 to 150000000;
output;
end;
run;quit;
%macro mke020(obs,txt);
%let interval=%eval(150000000/&obs);
data spde.dim&obs;
retain key 0 dim&interval "&txt";
do key=1 to 150000000 by &interval;
output;
end;
run;quit;
%mend mke020;
%mke020(500 ,MALE);
%mke020(1000 ,VT );
%mke020(2000 ,OLD );
%mke020(4000 ,BLUE);
%mke020(8000 ,TOP );
%mke020(10000 ,LOW );
%mke020(20000 ,HIGH);
%mke020(40000 ,PRES);
%mke020(80000 ,BEST);
%mke020(100000 ,GROW);
%mke020(200000 ,POOR);
%mke020(400000 ,RICH);
%mke020(800000 ,KID );
%mke020(1000000 ,POOL);
%mke020(2000000 ,MEAN);
%mke020(4000000 ,MIN );
%mke020(8000000 ,MAX );
%mke020(10000000 ,OUT );
%mke020(20000000 ,DEM );
%mke020(30000000 ,GOP );
data master;
merge spde.mm150(in=master)
spde.dim500
spde.dim1000
spde.dim2000
spde.dim4000
spde.dim8000
spde.dim10000
spde.dim20000
spde.dim40000
spde.dim80000
spde.dim100000
spde.dim200000
spde.dim400000
spde.dim800000
spde.dim1000000
spde.dim2000000
spde.dim4000000
spde.dim8000000
spde.dim10000000
spde.dim20000000
spde.dim30000000
spde.dim10000000
spde.dim20000000
spde.dim30000000
;
by key;
if master;
if mod(_n_,10000000)=0 then put _n_=;
run;quit;
NOTE: There were 150000000 observations read from the data set SPDE.MM150.
NOTE: There were 500 observations read from the data set SPDE.DIM500.
NOTE: There were 1000 observations read from the data set SPDE.DIM1000.
NOTE: There were 2000 observations read from the data set SPDE.DIM2000.
NOTE: There were 4000 observations read from the data set SPDE.DIM4000.
NOTE: There were 8000 observations read from the data set SPDE.DIM8000.
NOTE: There were 10000 observations read from the data set SPDE.DIM10000.
NOTE: There were 20000 observations read from the data set SPDE.DIM20000.
NOTE: There were 40000 observations read from the data set SPDE.DIM40000.
NOTE: There were 80000 observations read from the data set SPDE.DIM80000.
NOTE: There were 100000 observations read from the data set SPDE.DIM100000.
NOTE: There were 200000 observations read from the data set SPDE.DIM200000.
NOTE: There were 400000 observations read from the data set SPDE.DIM400000.
NOTE: There were 802140 observations read from the data set SPDE.DIM800000.
NOTE: There were 1000000 observations read from the data set SPDE.DIM1000000.
NOTE: There were 2000000 observations read from the data set SPDE.DIM2000000.
NOTE: There were 4054055 observations read from the data set SPDE.DIM4000000.
NOTE: There were 8333334 observations read from the data set SPDE.DIM8000000.
NOTE: There were 10000000 observations read from the data set SPDE.DIM10000000.
NOTE: There were 21428572 observations read from the data set SPDE.DIM20000000.
NOTE: There were 30000000 observations read from the data set SPDE.DIM30000000.
NOTE: There were 10000000 observations read from the data set SPDE.DIM10000000.
NOTE: There were 21428572 observations read from the data set SPDE.DIM20000000.
NOTE: There were 30000000 observations read from the data set SPDE.DIM30000000.
NOTE: The data set WORK.UNDER1MM has 150000000 observations and 35 variables.
NOTE: DATA statement used (Total process time):
real time 14:18.11
cpu time 19:15.79
127 ! quit;
```

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

11-17-2016 03:17 PM

@rogerjdeangelis even if agree on the overall conclusion I don't think that you can use your example as a template. Fact tables doesn't have a single FK column shated by all dimensions. This means that you can never have merge by more than one dimension at the time (different by variables).

Data never sleeps

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

11-17-2016 03:53 PM

It was a dumb example but does a lot of I/O.

I suspect you can use SQL left joins using mutiple pk and fk realtions and the performance would not be an oder of magnitude less than 25hrs

Generally there is an fk relation with dimension tables.

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

11-17-2016 04:07 PM

Correction

The performance WOULD be much less than 20hrs

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

11-17-2016 04:49 PM

To put a note to the whole chat..don't forget your friends DROP & KEEP and minimize the variables you mix and match.