256 threads maximum?

[Panagiotis Issaris]

Scott Lamb wrote:

>> The following simple testprogram tries to create the number of 
>> threads given as argv[1].
>> When I'm using a 2.6.3 kernel with glibc 2.3.2/NPTL 0.60, the maximum 
>> number of threads this application can create is 255. With any number 
>> higher it complains about memory allocation failing.
>
>
> I bet you're using a 32-bit machine with an 8 MiB stack. The top bit 
> is reserved for kernel use, IIRC, so a program can only address 2 GiB 
> of virtual memory. 2 GiB / 8 MiB = 256. One thread already exists. If 
> you ran a
>
>     $ ulimit -s 4096
>     $ ./thr 511
>
> It would succeed, I'd think. And 512 would fail.

Thanks!!! I validated your explanation, and it worked out just as you 
predicted!

> I'm not sure why LinuxThreads was able to allocate more threads. Maybe 
> it determines the stack size in a different way?
>
>> What am I doing wrong? I know that NPTL can do a lot better then 
>> this, since I have been reading reports on NPTL creating thousands of 
>> threads ( I think I recall someone stating creating 300000 threads 
>> from one process).
>
>
> You can do a little better by decreasing the stack space per thread, 
> as mentioned above. But I think they must have been doing this test on 
> a 64-bit machine.

Yes, you are absolutely right. Or they could have used a 7158 byte stack 
on a 32-bit machine :-)

Thanks for your explanation,
With friendly regards,
Takis