Very Computer

Hi, all

I have one process which forks many child processes.
how many child processes can parent process fork?
if there is a maximun which is less than what i expect, i'd better have
several same parent processes.
is that right?

and, one more question.
which one is more effective?
1. only one process doing a job.
2, several processes doing a job concurrently.

Hi Jaguk Ku,

there is usually a limit of "maximum processes per user" which can be
configured in
the Unix-kernel. This limit applies when forking children from one or from
more than one parents. (The only user without such a limit is root. The only
limit applying to root is "maximum processes in the system".)

Efficiency (Questions 1 and 2) is hard to tell, without knowing what you
want to do.
- Scenario-A) A process allocates large amounts of heap-memory...
   Such a process will run more efficiently if it serves more than one
   clients. Because multiple instances of such a process might require
   heavy swapping activities.

- Scenario-B) A process with small or medium requirements of heap-memory,
   but a number of blocking sequence points (such as database-transactions)
   or blocking calls to some library functions...
   Such a process might run better, if started unique for each client.

Regards Rainer

In theory, all of them. <g> (Think of init (pid 1) and how every other
process is somehow a decendant of init).

In practice, it's limited by various means, including quotas/ulimits, OS
implementation (max processes in process table), and OS design
(hardcoded fork() throttling logic).  There will also be exceptions and
escapehatches to these limits.

Quote:> if there is a maximun which is less than what i expect, i'd better have
> several same parent processes.

Which, of course, are children of another process <g>

Quote:> is that right?

It may be so.

Quote:> and, one more question.
> which one is more effective?
> 1. only one process doing a job.
> 2, several processes doing a job concurrently.

It depends on the job to be done. If the job is atomic (can't be
subdivided) then a single process will likely be more effective than
multiple processes because you'll avoid artifically imposed deadlocks
and race conditions. If the job can be subdivided effectively, then
several processes _may_ be more effective (assuming that there are
little or no external costs, like reduced processing intervals because
of the increased number of processes)

--
Lew Pitcher

Master Codewright and JOAT-in-training
Registered Linux User #112576