I will have look at POSIX::RT::Spawn

It is weird that there is no easy solution for this, is this also with python or other languages?
Because the mother process doesn't need to interact with sister process. I can write the information I need for 'blastn' to a file, then generate a new file that the mother process opens to read the result.<
So only the timing when the sister process starts is important, there is no need for a direct interaction.

It is weird that there is no easy solution for this, is this also with python or other languages?

Yes, any huge process trying to call fork() on this system will have the same problem. Well, I don't know that for sure, but we can check. Try to run this C program on the CentOS system:

/*
 * Pass this program the number of gigabytes it should
 * allocate before forking as the only command line argument
 * (fractions are accepted). For example: ./a.out 250
 * It will print any errors it encounters.
 */
#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>

int main(int argc, char ** argv) {
    void *ptr;

    if (argc != 2) return -1;
    ptr = malloc(1024 * 1024 * 1024 * atof(argv[1]));
    if (!ptr) {
        perror("malloc");
        return -2;
    }

    if (fork() < 0) perror("fork");

    free(ptr);
    return 0;
}
[download]

(To compile and run the program, try gcc program.c -o program && ./program 250 or even make program && ./program 250.)

Unless the language uses vfork() or posix_spawn() in its system() implementation, the call will fail. Perhaps the administrators of the CentOS machine could shed some light on the problem? They might know what to do better than me if you tell them that your processes have trouble forking when they allocate more than 50% RAM.