How to rightly use proc ds2 multiple threads to speed up data step?

[whymath]

I am learning to write PROC DS2, some materials says there is a threads= option, which enables SAS use multiple threads to run program, so SAS can run faster. I write the following program:

data test1;
  set sashelp.class;
  do i=1 to 3e6;
    output;
  end;
run;

*Normal data step;
data test2;
  set test1;
  sum_age+age;
run;

proc delete data=test2;
run;

*Data step in DS2, without multiple threads;
proc ds2;
  data test3/overwrite=yes;
    method run();
      dcl int sum_age;
      set test1;
      sum_age+age;
    end;
  enddata;
  run;
quit;

proc delete data=test3;
run;

*Data step in DS2, with multiple threads;
proc ds2;
  thread t/overwrite=yes;
    method run();
    dcl int sum_age;
    set test1;
    sum_age+age;
    end;
  endthread;
  run;

  data test4/overwrite=yes;
    dcl thread t t_instance;
    method run();
      set from t_instance threads=8;
    end;
  enddata;
  run;
quit;

Well, I just can't understand the running time:

Normal data step
real time: 2.97 seconds
cpu time: 2.87 seconds

Data step in DS2, without multiple threads
real time: 6.83 seconds
cpu time: 8.06 seconds

Data step in DS2, with multiple threads
real time: 4.47 seconds
cpu time: 13.26 seconds

And I write proc delete before taking each test, to keep the WORK library clean. My questions are:

1. Why data step in ds2 without multiple threads costs much time than normal data step?

2. Why data step in ds2 with multiple threads  costs much time than normal data step? Even I let threads=16, there is not significant difference on real time.

[AhmedAl_Attar · Accepted solution]

Hi @whymath 

Check out this paper "User-Defined Multithreading with the SAS® DS2 Procedure: Performance Testing DS2 Against Functionall..."

by Troy Martin Hughes 

He does an excellent job in explaining what you are observing.

Hope this helps,

Ahmed