Skip to contentSkip to conversation

2.4.x, mlockall() and pthreads: exceptionally huge memory demands

DISCUSSION Dan Kegel 1 participant 1 message Sep 10, 2001

lore
Share & tools
lore

Dan Kegel

author
#0 lore raw
Quoted 10 lines · [email protected] wrote:

I have written an application that uses mlockall() in conjunction with posix-threads. I realized that under Kernel 2.4.x (tested on 2.4.5-pre1 and 2.4.10-pre6, pentium III, no swap) each thread consumes an exceptionally huge amount of memory (about 2MB). When I omit the mlockall()-call each thread consumes only about 24KB.
Running the same application under kernel 2.2.x each thread consumes about 12KB of memory (with memory locked).
The attached program illustrates this behavior.

Quoted 2 lines

May this be a bug in the mlockall() implementation of 2.4.x and is this behavior already known (I found no hint in the archive)?

First - this is a glibc issue, not a kernel issue, I bet. Second - the new behavior seems allowed by the standard, if not desirable. Portably speaking, the default stack size for a thread may be quite large. If you want to consume less memory, either configure the threads to use a smaller stack with pthread_attr_setstacksize() (I haven't tried this), or use fewer threads (my favorite approach).

Third - the old behavior is hinted by
http://pauillac.inria.fr/~xleroy/linuxthreads/faq.html
(see below) so perhaps there's a problem with newer glibc's?
- Dan

"E.5: Does LinuxThreads implement pthread_attr_setstacksize() and pthread_attr_setstackaddr()?

These optional functions are provided in recent versions of LinuxThreads (0.8 and up). Earlier releases did not provide these optional components of the POSIX standard.

Even if pthread_attr_setstacksize() and pthread_attr_setstackaddr() are now provided, we still recommend that you do not use them unless you really have strong reasons for doing so. The default stack allocation strategy for LinuxThreads is nearly optimal:
stacks start small (4k) and automatically grow on demand to a fairly large limit (2M).
Moreover, there is no portable way to estimate the stack requirements of a thread, so setting the stack size yourself makes your program less reliable and non-portable."