topic Re: proc sql join efficiency in SAS Procedures

proc sql join efficiency

Steelers_In_DC — Fri, 28 Aug 2015 14:35:07 GMT

Good Morning All,

I have the following code:

proc sql;

create table fourth_run as

select b.*

from mdj.zip5_august a left join

mdj.addr_100m_zip9_zip5 b on

a.zip5 = substr(b.prop_zip_code,1,5);

zip5_august has about 900 records, 1 variable. addr_100m_zip9_zip5 has about 1 million records with 70 variables. I want to make sure that the way I'm doing this is best practice.

Is it best practice to have the smaller dataset listed first? Both are sorted accordingly, will the substr() cause much more processing time? Would it be better to set up another field? It's a monthly file that doesn't get used much. I'd prefer not to make too many changes as this server is fairly bogged down already.

Any input or suggestions are welcome.

Thanks,

Go Pirates.

Re: proc sql join efficiency

LinusH — Fri, 28 Aug 2015 15:11:54 GMT

I would index prop_zip_code.

Use options msglevel =i; to verify it's being used.

Re: proc sql join efficiency

Steelers_In_DC — Fri, 28 Aug 2015 15:49:31 GMT

I'll have to look that up. I'm not familiar with msglevel = i. I ran a small subset of each dataset and see this:

NOTE: SAS threaded sort was used.

I'm not sure what that means. Is an index necessary if it is sorted on prop_zip_code?

Re: proc sql join efficiency

LinusH — Fri, 28 Aug 2015 16:06:59 GMT

Not necessary, but you asked for optimzation, and an indexed join will usually perform better than a sort/merge join. Especially when the ratio of hits/total rows is as low as in your example.

The message means that SAS could use multi threading (that sort is done in parallel using multiple cores/CPUs). Good, but not surprising.

Re: proc sql join efficiency

Steelers_In_DC — Fri, 28 Aug 2015 16:27:26 GMT

Excellent. Thanks!