Very Computer

I'm looking into how often UNIX process ids (PIDs) are reused.

According to Leffler et al., "BSD maintains a range of unallocated PIDs
between [two values] mpid and pidchecked."

So on BSD, quickly allocated PIDs tend to increase by 1 for a time,
then jump up a bit and increase by 1 again.  Eventually (around 29999
it seems - maybe a configurable value?) they wrap around and start
increasing again. This system seems to be used in SVR4 as well.

What this means is that to get a process id reused you have to have a
busy machine.

Now, guess what? AIX has its own algorithm. It seems that PIDs on AIX
often increase by 255. Sometimes successive PIDs decrease in value for
a while. You never seem to get two consecutive integers as PIDs.

So I'm interested in any insight into how (and why) PIDs are allocated
in AIX, or other UN*Xes.

Thanks

-Andrew

--

Ingres, an ASK company, 1080 Marina Village Parkway, Alameda, California 94501
We're supposed to be XXXXing professionals (Mr. Pink)

Andrew> Now, guess what? AIX has its own algorithm. It seems that PIDs on AIX
Andrew> often increase by 255. Sometimes successive PIDs decrease in value for
Andrew> a while. You never seem to get two consecutive integers as PIDs.

Andrew> So I'm interested in any insight into how (and why) PIDs are allocated
Andrew> in AIX, or other UN*Xes.

This question pops up now and then. Most popular speculations are that
the Pentagon, the CIA or however didn't want consecutive pids because
that would be a security problem.

The truth, however, can be found in #include <sys/proc.h>. Internally
the pid consists of two parts: the slot in the process table and a
consecutivly assigned usage number within that slot. This makes access
to the process table a straight indexing operation, so efficiency
seems to be the real reason doing it different than others.
--