Setting up a performace test C vs awk?

I was told someone's awk code ran almost as fast as C. Not sure that's close to a true statement even in simple tasks. I wrote two tests and ran them with Linux's time command. Is this a fair and valid test? I need to give my management an honest answer.

Note that the code runs but I have to hand transcribe it.

### C

Code:

---

#include <stdio.h>
int main(void)
{
  FILE *fp;
  fp = fopen("/dev/null", "w");
  char my_string[10] = "123rte";
  int i;
  for ( i = 0; i < 10000; i++ )
  {
    fprintf(fp, "%s\n", my_string);
  }
  fclose(fp);
  return 0;
}

---

### awk

Code:

---

#!/bin/bash

my_string="123rte"
for (( n=0; n<10000; n++)) do
  echo $my_string | awk '{ print $1 }' > /dev/null 2>&1
done

---

Forgot to mention that the C version takes ~ 0.004 seconds and the awk version takes ~20 seconds.

No, that's not a fair test. When you do a "man awk", you will see that there are many complex things you can do with it by calling it just once. The reason your example is so slow is because it does basically nothing and the overhead from so many processes being created and destroyed is enormous.

Try this instead, which is more or less equivalent to the C code:

Code:

---

#!/bin/bash
 
my_string="123rte"
yes $my_string | head -n 10000 | awk '{ print $1 }' > /dev/null 2>&1

---

The equivalent awk code would be

Code:

---

$ cat foo.c
#include <stdio.h>
int main(void)
{
  FILE *fp;
  fp = fopen("/dev/null", "w");
  char my_string[10] = "123rte";
  int i;
  for ( i = 0; i < 1000000; i++ )
  {
    fprintf(fp, "%s\n", my_string);
  }
  fclose(fp);
  return 0;
}
$gcc foo.c
$ time ./a.out

real        0m0.113s
user        0m0.112s
sys        0m0.000s
$ cat foo.awk
#!/bin/awk
BEGIN {
  for ( i = 0; i < 1000000; i++ )
    print "123rte";
}
$ time awk -f foo.awk > /dev/null

real        0m0.140s
user        0m0.136s
sys        0m0.000s

---

Yes, awk is almost as fast.

But part of the attraction is being able to do regular expressions directly in the language, instead of having to do an awful lot of code yourself (even with a library).

Saw your first note and, with help from Google, came up with:

Code:

---

awk 'BEGIN{for(i=0;i<10000;i++) printf "123rte" > dev/null; }'

---

Which runs in 0.023 seconds. Comparable to python (0.026s), slower than C (0.004s), much faster than bash (0.414s) and naturally faster than the bash to awk handoff (18.566s)

Thank you!

While it's much faster than the combined version I'm still getting 0.023s (awk) vs 0.004s (C). Still, that's good info to pass on, I don't want to misrepresent.

It stands to reason that its speed would be comparable to python, since both are interpreted languages.

Quote:

---

Originally Posted by Leam

I was told someone's awk code ran almost as fast as C.

...

Code:

---

#!/bin/bash

my_string="123rte"
for (( n=0; n<10000; n++)) do
  echo $my_string | awk '{ print $1 }' > /dev/null 2>&1
done

---

---

The problem with this benchmark is that you're launching awk 10000 times to do a very small amount of work. I would imagine most of the execution time of this would be spent creating and destroying subprocesses.