Inspired by Instagram’s method to generate unique id’s. I just tried to write a program which does the same. Before that, why this? In the case of Instagram, they had their database sharded, with so many insertions in a second, each new record needs a unique id where databases auto increment doesn’t work and also place where you need to have id which are time sortable. The below is the program and it’s explanation.
#include<stdio.h>
#include<sys/time.h>
int main()
{
int i;
for(i =0;i < 32; i++)
func(i);
return 0;
}
int func(unsigned long u_id)
{
struct timeval t;
unsigned long id;
gettimeofday(&t,NULL);
id = (t.tv_sec * 1000 * 1000) + (t.tv_usec * 1000) << 42;
id |= (u_id % 16777216) << 24;
printf("%lu\n ",id);
return 0;
}
The main()
is self explanatory. func()
uses the gettimeofday function to get
the time in seconds and microseconds (I am not sure about the epoch), both the
values are converted in milliseconds. I have used 42 bits to save the
milliseconds which roughly gives more than 4 years of unique ids, now that we
have a time sortable id, we can add the last 22 bits with a unique id like a
database generated one or a user-id. But managing the last few bits with other
attributes solves the real purpose.
Site design and logo and content © fossix.org