centos 6.3 下分析程序虚拟内存偏高的问题

[lemonlinger]

两段程序，分别如下：
程序thread.c

#include <stdio.h>
#include <stdlib.h>
#include <pthread.h>

static void *
thread_start(void *arg)
{
	int s = 0;
	pthread_attr_t gattr;
	pthread_detach(pthread_self());
	/*
	s = pthread_getattr_np(pthread_self(), &gattr);
	*/

	while(1){
		usleep(100);
	} 
}

int
main(int argc, char *argv[])
{
	pthread_t thr;
	int s = 0;
	s = pthread_create(&thr, NULL, &thread_start, NULL);

	pause();
}

程序thread1.c

#include <stdio.h>
#include <stdlib.h>
#include <pthread.h>

static void *
thread_start(void *arg)
{
	int s = 0;
	pthread_attr_t gattr;
	pthread_detach(pthread_self());
	s = pthread_getattr_np(pthread_self(), &gattr);

	while(1){
		usleep(100);
	} 
}

int
main(int argc, char *argv[])
{
	pthread_t thr;
	int s = 0;
	s = pthread_create(&thr, NULL, &thread_start, NULL);

	pause();
}

在CentOS 6.2上分别编译成thread和thread1，运行，用top查看如下：

对thread，用pmap查看内存分配情况，如下：
这是pmap -x的结果：

这是pmap -d的结果：

对thread1，用pmap查看内存分配情况，如下：
这是pmap -x的结果：

这是pmap -d的结果：


从top命令的结果可以看出两者的虚拟内存相差将近5倍，可是代码仅仅差一行。
从两者的pmap -x的图里面比较，会发现，比较大的差别是thread1的内存中多了这个：


想请教一下两者的虚拟内存占用会差别这么大？

另外，用pmap查看的时候Mapping列的[anon]是代表什么意思？
还有，Mode列的----代表什么意思，为什么什么权限都没有？

谢谢啦~

[signforlin]

anon是已分配内存 stack是堆栈 mode：r=read, w=write, x=execute, s=shared, p=private。 获取本线程属性这个，还得大牛来看看。

[qq120848369]

When the thread attributes object returned by pthread_getattr_np() is no longer required, it should be destroyed using pthread_attr_destroy(3)

[lemonlinger]

引用 2 楼 qq120848369 的回复:

When the thread attributes object returned by pthread_getattr_np() is no longer required, it should be destroyed using pthread_attr_destroy(3)

加上了，但是问题依旧。

[lemonlinger]

有没有高手来解决一下，这是我查到的资料，分享一下，看看有没有帮助。 http://www.linuxquestions.org/questions/linux-kernel-70/virt-memory-jumps-by-64m-900398/ http://bk.kolics.net/?p=57 http://www.quora.com/Why-do-some-applications-use-significantly-more-virtual-memory-on-RHEL-6-compared-to-RHEL-5

[lemonlinger]

引用 4 楼 lemonlinger 的回复:

有没有高手来解决一下，这是我查到的资料，分享一下，看看有没有帮助。 http://www.linuxquestions.org/questions/linux-kernel-70/virt-memory-jumps-by-64m-900398/ http://bk.kolics.net/?p=57 http://www.quora.com/Why-do-some-a……

答案就在这个链接中： http://www.quora.com/Why-do-some-applications-use-significantly-more-virtual-memory-on-RHEL-6-compared-to-RHEL-5 原因如下： The reason for this is that RHEL 6 includes glibc 2.11 which includes a new "per thread arena memory allocator." This allocator will map many anonymous heaps and try to claim them to be thread-specific. On 64-bit systems, it defaults to allowing up to 8*number_of_cores arenas, and each will be 64M big. This will lead to an extra 4G of virtual memory used for an application with a lot of threads (eg Java servers). 解决办法： This behavior can be somewhat constrained by setting MALLOC_ARENA_MAX=n to set a maximum number of malloc arenas to allocate. 值得注意的是： Note that this is all virtual memory, not committed/resident memory. An application should not use appreciably more resident memory than it did under RHEL 5.