2009-06-08 13:39:47     Complete memory accounting

Document created by Aaronwu Employee on Aug 15, 2013
Version 1Show Document
  • View in full screen mode

2009-06-08 13:39:47     Complete memory accounting

John Goyette (UNITED STATES)

Message: 75367   

 

I would like to understand how to get a complete picture of how much RAM is being used by the kernel and user space applications.  I have found that 'ps', 'free' and '/proc/meminfo' do not provide a full picture.  What other tools and methods to you recommend for capturing this information? 

 

I have built a minimal distribution for the 527 EZ-Kit and booted it with an NFS mounted root file system.  I told the kernel via command line that it has 32 Mb of RAM.  Searching the dmesg output, I can find what the kernel is reporting as follows:

 

root:/> dmesg | grep -i memory

Board Memory: 32MB

Kernel Managed Memory: 32MB

Memory map:

Memory available: 29320k/32768k RAM, (100k init code, 1376k kernel code, 640k data, 1024k dma, 304k reserved)

Freeing unused kernel memory: 100k freed

 

And the 'free' command produces the following:

 

root:/> free

              total         used         free       shared      buffers

  Mem:        29420         3948        25472            0            0

 

 

This tells me that of my 32 Mb of RAM, the kernel is using 3,344k leaving 29,420k available which is reported as "total" in the free command.  This, I think, is the easy part.  The difficulty comes in determining how the remaining 29,420k is used.

 

The 'ps' command produces the following output:

 

root:/> ps

  PID USER       VSZ STAT COMMAND

    1 root       108 S    /sbin/init

    2 root         0 SW<  [kthreadd]

    3 root         0 SW<  [ksoftirqd/0]

    4 root         0 SW<  [events/0]

    5 root         0 SW<  [khelper]

   47 root         0 SW<  [kblockd/0]

   74 root         0 SW   [pdflush]

   75 root         0 SW   [pdflush]

   76 root         0 SW<  [kswapd0]

   77 root         0 SW<  [aio/0]

   78 root         0 SW<  [nfsiod]

  128 root         0 SW<  [bfin-spi.0]

  138 root         0 SW<  [hid_compat]

  141 root         0 SW<  [rpciod/0]

  155 root       532 S    -/bin/sh

  156 root        40 S    /sbin/inetd

  158 root       424 S    /sbin/klogd -n

  159 root       428 S    /sbin/syslogd -n

  167 root       428 R    ps

 

 

When you add up the running user space applications (not including ps itself) , you end up with 1,532k.  However, the free command and /proc/meminfo report that 3,948k is being used.  My question is, what is consuming the remaining roughly 2.5 Mb of RAM?  Is this being used by the kernel threads indicated in brackets in the ps command?  If so, how do I find out how much each of those is using?  Is it being used by RAMFS mounts (I have only /var and /tmp mounted as ramfs and tmpfs)?  What tools can we use to investigate where that memory is being spent? 

 

For reference, I am also including the contents of /proc/meminfo and the rc file:

 

 

root:/> cat /proc/meminfo

MemTotal:          29420 kB

MemFree:           25456 kB

Buffers:               0 kB

Cached:              560 kB

SwapCached:            0 kB

Active:              440 kB

Inactive:            120 kB

Active(anon):          0 kB

Inactive(anon):        0 kB

Active(file):        440 kB

Inactive(file):      120 kB

SwapTotal:             0 kB

SwapFree:              0 kB

Dirty:                 0 kB

Writeback:             0 kB

AnonPages:             0 kB

Mapped:                0 kB

Slab:                752 kB

SReclaimable:         84 kB

SUnreclaim:          668 kB

PageTables:            0 kB

NFS_Unstable:          0 kB

Bounce:                0 kB

WritebackTmp:          0 kB

CommitLimit:       14708 kB

Committed_AS:          0 kB

VmallocTotal:          0 kB

VmallocUsed:           0 kB

VmallocChunk:          0 kB

 

 

 

root:/> cat /etc/rc

hostname blackfin

mount -t proc proc /proc -o noexec,nosuid,nodev

mount -t sysfs sysfs /sys -o noexec,nosuid,nodev

mkdir -p /dev/pts

mount -t devpts devpts /dev/pts -o noexec,nosuid

mount -t ramfs var /var

mkdir /var/tmp /var/log /var/run /var/lock

mount -t tmpfs tmp /tmp -o nosuid,nodev

 

Any insight would be appreciated.  Thanks.

 

-John G.

 

QuoteReplyEditDelete

 

 

2009-06-08 16:26:30     Re: Complete memory accounting

Robin Getz (UNITED STATES)

Message: 75370   

 

John:

 

Anything with a cached based file system is going to be a little hard to tell what is going on - should you include the cache as "free" memory (since it can be dumped and freed for the application - or "used" memory - since it is in use.

 

/proc/meminfo tells you what things are going on, and let's you make up the decision...

 

MemTotal is the total memory avalible to userspace and the kernel.

 

Cached is the file system cache.

 

Slab is the kernel dynamic memory (kmalloc, and friends) + userspace (since that is how userspace gets it's memory).

 

Does that help?

QuoteReplyEditDelete

 

 

2009-06-08 22:09:40     Re: Complete memory accounting

John Goyette (UNITED STATES)

Message: 75374   

 

Robin,

 

Thanks, that is helpful.  Now I understand /proc/meminfo a little better.  So, I can account for another 1.2 Mb between the cache and slab.  I suspect that the rest of the RAM is used by my /tmp and /var file systems and the 'cat' program running at the time I view /proc/meminfo.  Kind of funny to think of it that way, but if you type cat /proc/meminfo, the MemFree entry must reflect the fact that you have an instance of busybox running that is executing cat, right?

 

Is there anyway for me to see how much RAM is in use for /tmp and /var?  Since those are ramfs, they are not cached, correct? The du command gives the size of the filesystem, but is the kernel using more RAM than that for the filesystem?

 

Thanks again for your help.

 

-John G.

QuoteReplyEditDelete

 

 

2009-06-08 22:26:04     Re: Complete memory accounting

Robin Getz (UNITED STATES)

Message: 75375   

 

John:

 

It depends on the backing file system. There are a few - ramfs lives in the page cache - so it should be accounted for in the page cache accounting.

 

  wiki.debian.org/ramfs

 

ram disks are different. - (see the wiki above). The answer is - it depends.

 

-Robin

Attachments

    Outcomes