Linux进程虚拟内存和物理内存

先介绍几个基本概念:

SIZE: 进程使用的地址空间, 如果进程映射了100M的内存, 进程的地址空间将报告为100M内存. 事实上, 这个大小不是一个程序实际使用的内存数.

RSS: "Resident Set Size", 实际驻留"在内存中"的内存数. 不包括已经交换出去的代码. 举一个例子: 如果你有一个程序使用了100K内存, 操作系统交换出40K内存, 那么RSS为60K. RSS还包括了与其它进程共享的内存区域. 这些区域通常用于libc库等.

SHARE: RSS中与其它进程共享的内存部分大小.

VMSIZE: 一个进程占用的总的地址空间大小. 它包括了没有映射到内存中的页面.

Private RSS: 映射到内存中的页面, 这些页面仅由进程单独使用. 这也是我们最关心地方: 进程实际占用的内存数.

如何来查看Private RSS呢? /proc接口中每一个进程目录下的smaps提供了private rss信息. smaps是在2.6.16内核版本引进来的.

私有驻留内存数(Private RSS):

查看/proc/$pid/smaps

下面我以本站使用的Linux AS5为例, 查看进程号1293(vmware-guestd, 本站使用的是一台基于vmware的客户虚拟机,)使用的Private RSS数.

CODE:

# cat /proc/1293/smaps

00111000-00112000 rwxp 00111000 00:00 0

Size: 4 kB

Rss: 4 kB

Shared_Clean: 0 kB

Shared_Dirty: 0 kB

Private_Clean: 0 kB

Private_Dirty: 4 kB

0050e000-0050f000 rwxp 0050e000 00:00 0

Size: 4 kB

Rss: 0 kB

Shared_Clean: 0 kB

Shared_Dirty: 0 kB

Private_Clean: 0 kB

Private_Dirty: 0 kB

0051a000-0051b000 r-xp 0051a000 00:00 0 [vdso]

Size: 4 kB

Rss: 4 kB

Shared_Clean: 4 kB

Shared_Dirty: 0 kB

Private_Clean: 0 kB

Private_Dirty: 0 kB

0051b000-00534000 r-xp 00000000 fd:00 194898 /lib/ld-2.4.so

Size: 100 kB

Rss: 0 kB

Shared_Clean: 0 kB

Shared_Dirty: 0 kB

Private_Clean: 0 kB

Private_Dirty: 0 kB

00534000-00535000 r-xp 00018000 fd:00 194898 /lib/ld-2.4.so

Size: 4 kB

Rss: 0 kB

Shared_Clean: 0 kB

Shared_Dirty: 0 kB

Private_Clean: 0 kB

Private_Dirty: 0 kB

00535000-00536000 rwxp 00019000 fd:00 194898 /lib/ld-2.4.so

Size: 4 kB

Rss: 0 kB

Shared_Clean: 0 kB

Shared_Dirty: 0 kB

Private_Clean: 0 kB

Private_Dirty: 0 kB

00538000-00665000 r-xp 00000000 fd:00 194905 /lib/libc-2.4.so

Size: 1204 kB

Rss: 212 kB

Shared_Clean: 204 kB

Shared_Dirty: 0 kB

Private_Clean: 8 kB

Private_Dirty: 0 kB

00665000-00667000 r-xp
0012d000 fd:00 194905 /lib/libc-2.4.so

Size: 8 kB

Rss: 8 kB

Shared_Clean: 4 kB

Shared_Dirty: 0 kB

Private_Clean: 0 kB

Private_Dirty: 4 kB

00667000-00668000 rwxp
0012f000 fd:00 194905 /lib/libc-2.4.so

Size: 4 kB

Rss: 4 kB

Shared_Clean: 0 kB

Shared_Dirty: 0 kB

Private_Clean: 0 kB

Private_Dirty: 4 kB

00668000-0066b000 rwxp 00668000 00:00 0

Size: 12 kB

Rss: 8 kB

Shared_Clean: 0 kB

Shared_Dirty: 0 kB

Private_Clean: 0 kB

Private_Dirty: 8 kB

08047000-08062000 r-xp 00000000 fd:00 292327 /usr/sbin/vmware-guestd

Size: 108 kB

Rss: 64 kB

Shared_Clean: 0 kB

Shared_Dirty: 0 kB

Private_Clean: 64 kB

Private_Dirty: 0 kB

08062000-08063000 rwxp
0001a000 fd:00 292327 /usr/sbin/vmware-guestd

Size: 4 kB

Rss: 4 kB

Shared_Clean: 0 kB

Shared_Dirty: 0 kB

Private_Clean: 0 kB

Private_Dirty: 4 kB

08063000-08068000 rwxp 08063000 00:00 0

Size: 20 kB

Rss: 12 kB

Shared_Clean: 0 kB

Shared_Dirty: 0 kB

Private_Clean: 0 kB

Private_Dirty: 12 kB

08385000-08886000 rwxp 08385000 00:00 0 [heap]

Size: 5124 kB

Rss: 5080 kB

Shared_Clean: 0 kB

Shared_Dirty: 0 kB

Private_Clean: 0 kB

Private_Dirty: 5080 kB

bfeb2000-bfec7000 rwxp bfeb2000 00:00 0 [stack]

Size: 84 kB

Rss: 12 kB

Shared_Clean: 0 kB

Shared_Dirty: 0 kB

Private_Clean: 0 kB

Private_Dirty: 12 kB

上面我们看到从smaps看不太方便, 推荐使用Ben Maurer写的perl脚本:

#!/usr/bin/perl

# Copyright Ben Maurer
# you can distribute this under the MIT/X11 License

use Linux::Smaps;

my $pid=shift @ARGV;
unless ($pid) {
print "./smem.pl <pid>\n";
exit 1;
}
my $map=Linux::Smaps->new($pid);
my @VMAs = $map->vmas;

format STDOUT =
VMSIZE:  @######## kb
$map->size
RSS:     @######## kb total
$map->rss
@######## kb shared
$map->shared_clean + $map->shared_dirty
@######## kb private clean
$map->private_clean
@######## kb private dirty
$map->private_dirty
.

write;

printPrivateMappings ();
printSharedMappings ();

sub sharedMappings () {
return grep { ($_->shared_clean  + $_->shared_dirty) > 0 } @VMAs;
}

sub privateMappings () {
return grep { ($_->private_clean  + $_->private_dirty) > 0 } @VMAs;
}

sub printPrivateMappings ()
{
$TYPE = "PRIVATE MAPPINGS";
$^ = 'SECTION_HEADER';
$~ = 'SECTION_ITEM';
$- = 0;
$= = 100000000;
foreach  $vma (sort {-($a->private_dirty <=> $b->private_dirty)}
privateMappings ()) {
$size  = $vma->size;
$dirty = $vma->private_dirty;
$clean = $vma->private_clean;
$file  = $vma->file_name;
write;
}
}

sub printSharedMappings ()
{
$TYPE = "SHARED MAPPINGS";
$^ = 'SECTION_HEADER';
$~ = 'SECTION_ITEM';
$- = 0;
$= = 100000000;

foreach  $vma (sort {-(($a->shared_clean + $a->shared_dirty)
<=>
($b->shared_clean + $b->shared_dirty))}
sharedMappings ()) {

$size  = $vma->size;
$dirty = $vma->shared_dirty;
$clean = $vma->shared_clean;
$file  = $vma->file_name;
write;

}
}

format SECTION_HEADER =
@<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
$TYPE
@>>>>>>>>>> @>>>>>>>>>>  @>>>>>>>>>   @<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
"vmsize" "rss clean" "rss dirty" "file"
.

format SECTION_ITEM =
@####### kb @####### kb @####### kb   @<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
$size $clean $dirty $file
.

下面是由seme.pl脚本解析的smaps数据:

首先安装Linux::Smaps模块:

perl -MCPAN -e 'install Linux::Smaps'//需要网络，如无法上网，则可下载压缩包解压到 /root

然后用seme.pl解析1293进程的Smaps数据:

CODE:

# ./smem.pl 1293

VMSIZE: 7200 kb

RSS: 1052 kb total

192 kb shared

100 kb private clean

760 kb private dirty

PRIVATE MAPPINGS

vmsize rss clean rss dirty file

5636 kb 8 kb 724 kb [heap]

84 kb 0 kb 12 kb [stack]

4 kb 0 kb 4 kb

8 kb 0 kb 4 kb
/lib/libc-2.4.so

4 kb 0 kb 4 kb
/lib/libc-2.4.so

12 kb 4 kb 4 kb

4 kb 0 kb 4 kb
/usr/sbin/vmware-guestd

20 kb 8 kb 4 kb

1204 kb 16 kb 0 kb
/lib/libc-2.4.so

108 kb 64 kb 0 kb
/usr/sbin/vmware-guestd

SHARED MAPPINGS

vmsize rss clean rss dirty file

1204 kb 188 kb 0 kb
/lib/libc-2.4.so

4 kb 4 kb 0 kb [vdso]

从上面看到rss大小被分成了两个部分: private(私有)和shared(共享).

private rss就是我们最关心的进程实际占用的内存数.