Is there any limitations to number of running processes on host? - process

Question

In platform file I have only one host:
<host id="Worker1" speed="100Mf" core="101"/>
Then in worker.c I create 101 (or > 100) processes expecting that each on each core one process will be launched. But I noticed that only 100 first processes able to execute task or write with XBT_INFO:
int worker(int argc, char *argv[])
{
for (int i = 0; i < 101; ++i) {
MSG_process_create("x", slave, NULL, MSG_host_self());
}
return 0;
}
int slave(){
MSG_task_execute(MSG_task_create("kotok", 1e6, 0, NULL));
MSG_process_kill(MSG_process_self());
return 0;
}
Other processes above 100 first ones are unable to manage and kill:
[ 1.000000] (0:maestro#) Oops ! Deadlock or code not perfectly clean.
[ 1.000000] (0:maestro#) 1 processes are still running, waiting for something.
[ 1.000000] (0:maestro#) Legend of the following listing: "Process <pid> (<name>#<host>): <status>"
[ 1.000000] (0:maestro#) Process 102 (x#Worker1): waiting for execution synchro 0x26484d0 (kotok) in state 2 to finish
UPDATE
Here some code functions are:
main
int main(int argc, char *argv[])
{
MSG_init(&argc, argv);
MSG_create_environment(argv[1]); /** - Load the platform description */
MSG_function_register("worker", worker);
MSG_launch_application(argv[2]); /** - Deploy the application */
msg_error_t res = MSG_main(); /** - Run the simulation */
XBT_INFO("Simulation time %g", MSG_get_clock());
return res != MSG_OK;
}
deployment.xml
<?xml version='1.0'?>
<!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid/simgrid.dtd">
<platform version="4">
<process host="Worker1" function="worker">
<argument value="0"/>
</process>
</platform>

Answer 1

There is actually an internal limit for the size of a maxmin system (the core of SimGrid), which is 100, and may be hit in this case. I just added a flag to make this limit configurable. Could you pull the last version, and try setting maxmin/concurrency_limit to 1000 and see if it fixes your issue?

Answer 2

the number of process that can be started on a host has nothing to do with the number of cores. As on a real machine, you can have several processes running "simultaneously" thanks to time sharing mechanisms. It's the same here. when the number of running processes is greater than the number of cores (be it 1 or more), they have to share the resources.
The cause of your issue is elsewhere, but you do not provide a full minimal working example (main? deployment file?) and it's hard to help.