Re: speeding up a query

[Marcus Engene <mengpg2@xxxxxxxxx>]

Hi again,

I was thinking, in my slow query it seems the sorting is the villain. 
Doing a simple qsort test I notice that:
ehsmeng@menglap /cygdrive/c/pond/dev/tt
$ time ./a.exe 430

real    0m0.051s
user    0m0.030s
sys     0m0.000s

ehsmeng@menglap /cygdrive/c/pond/dev/tt
$ time ./a.exe 430000

real    0m0.238s
user    0m0.218s
sys     0m0.015s

ehsmeng@menglap /cygdrive/c/pond/dev/tt
$ time ./a.exe 4300000

real    0m2.594s
user    0m2.061s
sys     0m0.108s

From this very unfair test indeed I see that my machine has the 
capability to sort 4.3 million entries during the same time my pg is 
sorting 430.

And i cannot stop wondering if there is some generic sorting routine 
that is incredibly slow? Would it be possible to, in the situations 
where order by is by simple datatypes of one column, to do a special 
sorting, like the qsort example in the end of this mail? Is this already 
addressed in later versions?

If no, why? and if yes, where in the pg code do I look?

Best regards,
Marcus


#include <stdio.h>
#include <stdlib.h>

typedef struct {
   int     val;
   void   *pek;
} QSORTSTRUCT_INT_S;

int sortstruct_int_compare(void const *a, void  const *b)
{
   return ( ((QSORTSTRUCT_INT_S *)a)->val - ((QSORTSTRUCT_INT_S 
*)b)->val );
}

int main (int argc, char **argv)
{
   int nbr = 0;
   int i = 0;
   QSORTSTRUCT_INT_S *sort_arr = 0;
   if (1 == argc) {
       printf("forgot amount argument\n");
       exit(1);
   }
   nbr = atoi (argv[1]);
   if (0 == (sort_arr = malloc (sizeof(QSORTSTRUCT_INT_S) * nbr))) {
       printf("cannot alloc\n");
       exit(1);
   }
   srand(123);
   for (i=0; i<nbr; i++) {
       sort_arr[i].val = rand();
   }
   qsort(sort_arr, nbr, sizeof(QSORTSTRUCT_INT_S),sortstruct_int_compare);
   return 0;
}