Using SAS Enterprise Guide Effectively

Who Does What Where ?

Now What ?

EG Default Behaviour

EG Default Setup

EG SASUSER

EG Profiles

EG Profiles

EG AUTOEXEC --- kinda

EG Who Does What Where ?

different libraries different databases millions of rows SY tomorrow !

EG Who Does What Where ?

same library same database candidate for pass-through

What is Pass-Through ?
SAS Access engines for RDBMS eg. Teradata, Oracle, ODBC essentially APIs to communicate with the DB explicit pass-through o SAS program connects to DB o SQL is passed through to DB SQL must be native to DB implicit pass-through o SAS decides and takes care of it for you

Explicit Pass-Through
open a program window in EG
proc sql; connect to teradata ( user=&user pass="&pass" server=&server); create table pp_sum_cube as select * from connection to teradata ( select s.brand_name, m.* from mi_analytics.v_bm_f_pp_migration_sum_cube m left join mi_data.bm_d_segment s on m.segment_cd = s.segment_cd where m.period_dt in ( '2012-03-01', '2012-03-02' ) order by m.period_dt, s.brand_name ); quit;

Implicit Pass-Through
You write SAS SAS interprets your code and it writes Teradata SQL o functions o ... if possible ... In Database processing o SAS Procedures o SAS formats

Implicit Pass-Through
libname mitddata teradata user=&user pass="&pass" server=&server sql_functions=all ;

proc sql; select from group order ; quit;

segment_tier_0, count(*) as cnt mitddata.BM_D_Segment by segment_tier_0 by segment_tier_0

SAS LIBNAME statement Teradata engine SAS SQL summary query runs entirely in Teradata o small result set returned to SAS

Implicit Pass-Through
libname mitddata teradata user=&user pass="&pass" server=&server sql_functions=all; proc freq data = mitddata.BM_D_Segment; tables segment_tier_0 /missing nopercent ; run; 514 515 516 proc freq data = mitddata.BM_D_Segment; tables segment_tier_0 /missing nopercent ; run;

NOTE: SQL generation will be used to construct frequency and crosstabulation tables.

Implicit Pass-Through Uh oh...

proc sql; select adj_type, adj_reason, count(*) as cnt from mitddata.BM_D_DA_Adjustments where substr(adj_reason,1,3) = 'M2M' and intck('month',eff_dt,exp_dt) > 1 group by adj_type, adj_reason order by adj_type, adj_reason ; quit;

runs forevvvvvvvvvver

Implicit Pass-Through Uh oh...

SUBSTR and INTCK are SAS functions... SQL_Functions = ALL finite list SUBSTR is there not INTCK what did SAS do ? better than you might imagine options sastrace=',,,d' sastraceloc=saslog nostsuffix;

Implicit Pass-Through Uh oh...

SAS_SQL: Unable to convert the query to a DBMS specific SQL statement due to an error. ACCESS ENGINE: SQL statement was not passed to the DBMS, SAS will do the processing.
TERADATA: trqacol- No casting. Raw row size=66, Casted size=74, CAST_OVERHEAD_MAXPERCENT=20% TERADATA_8: Prepared: on connection 12

SELECT "ADJ_TYPE","ADJ_REASON","EFF_DT","EXP_DT" FROM mi_data."BM_D_DA_Adjustments" WHERE ( (SUBSTR("ADJ_REASON", 1, 3) = 'M2M' ) )

TERADATA_9: Executed: on connection 12 SELECT "ADJ_TYPE","ADJ_REASON","EFF_DT","EXP_DT" FROM mi_data."BM_D_DA_Adjustments" SUBSTR("ADJ_REASON", 1, 3) = 'M2M' ) )

WHERE

( (

TERADATA: trget - rows to fetch: 30

TERADATA: trforc: COMMIT WORK

Implicit Pass-Through Caveat

Due to incompatibility in date and time functions between Teradata and SAS, Teradata might not process them correctly. Check your results to determine whether these functions are working as expected how to be certain ? o explicit pass-through

Implicit Pass-Through Ummmm

Oracle DB o two schemas BMLAPPO BMLREF

SAS will happily run in Oracle turns SQL into Oracle query much happiness

Implicit Pass-Through Ummmm

Teradata DB o two schemas MI_DATA MI_STAGE

15,000 rows left joined to 120,000,000 o 576 amps !??! 48 minutes later...

Implicit Pass-Through NOT !

TERADATA_7: Prepared: on connection 6 SELECT * FROM mi_data."BM_D_SOC_Price_Plan

ERROR: This SQL statement will not be passed to the DBMS for processing because it involves a join across librefs with different connection properties.
TERADATA: trqacol- Casting decimals. Raw row size=38, Casted size=38, CAST_OVERHEAD_MAXPERCENT=20% TERADATA_8: Prepared: on connection 4 SELECT "OLD_PP_SOC_CD","NEW_PP_SOC_CD",CAST("SERVICE_ACTIVITY_KEY" AS FLOAT),"PERIOD_DT",CAST("SERVICE_ACTIVITY_KEY_PREV" AS FLOAT) FROM mi_stage."STG_SC_PP_MIGRATION"

NOTE: PROCEDURE SQL used (Total process time): 48:12.89 real time user cpu time 30:29.85 system cpu time 6:40.33

Explicit Pass-Through Joy

select * connection to teradata ( select .. from MI_STAGE.ODS_SC_DAILY_ACTIVITY, MI_STAGE.STG_SC_PP_MIGRATION left join MI_DATA.BM_D_SOC_Price_Plan as BM_D_SOC_Price_Plan_prev on STG_SC_PP_MIGRATION.OLD_PP_SOC_CD = BM_D_SOC_Price_Plan_prev.PP_SOC_CD where STG_SC_PP_MIGRATION.SERVICE_ACTIVITY_KEY = ODS_SC_DAILY_ACTIVITY.SERVICE_ACTIVITY_KEY and STG_SC_PP_MIGRATION.PERIOD_DT = ODS_SC_DAILY_ACTIVITY.PERIOD_DT and ODS_SC_DAILY_ACTIVITY.ADJ_KEY = 0 );

NOTE: PROCEDURE SQL used (Total process time): 4.66 seconds real time user cpu time 0.22 seconds system cpu time 0.04 seconds

Temporary Tables

intermediate results o subsetting exercise o deep dive analysis

need historical data for these 10,000 subscribers 1. Pull subscriber history down to SAS, join 2. where subscriber_no in ( 123,345,567 x 10K ) 3. Push 10,000 subscriber numbers up o do the join in DB o pull only what you need down o minimize slow data transfers

Temporary Tables
Oracle Teradata global temp tables volatile tables

exist for the duration of the session o like SAS WORK datasets requires o LIBNAME engine o explicit pass-thru EXECUTE ( DDL statement ) by teradata; o implicit pass-thru to load subset o explicit pass-thru for final result

Temporary Tables
libname td_volt teradata server="&server" user=&user password="&pass" dbmstemp=yes connection=global dbcommit=0; proc sql; connect to teradata ( server="&server" user=&user password="&pass" connection=global mode=teradata ); /* create an empty volatile table */ execute ( create multiset volatile table test_vol ... ( .. columns .. ) ) by teradata; quit;

Temporary Tables
/* Put rows into the volatile table via the libname if lots of records, additional options necessary, e.g. fastload */ proc sql; insert into td_volt.test_vol select subscriber_number as subscriber_no from midata.dim_subscriber ( obs = 100 ) ; quit;

Temporary Tables
/* Execute pass-thru query to join Teradata table to volatile table */ proc sql; connect to teradata (server="&server" user=&user password="&pass" connection=global); create table sas_datastet_of_results as select * from connection to teradata ( select s.* from bmbi_view.vb_subscriber s, test_vol v where v.subscriber_no = s.sub_no ); quit;

Conclusion
do stuff where it makes sense o use power of DB o summarize, subset, sort in DB not all implicit pass-thru is o use option SASTRACE use EXPLAIN

Contact
Harry Droogendyk harry.droogendyk@bell.ca harry@stratia.ca Phone: Web: 905-282-3476 www.stratia.ca/papers