/dev/shm "No space left on device" but df -h shows only 20% used.

amunro · 12-20-2011, 04:25 AM

I run centOS 5 on a VPS.

I am trying to write to a ramdisk at /dev/shm as my IO speed is terrible, however everytime I try and write to the directory, it copies around 700mb and then tells me there is no space left.

However, when I run df, it tells me that there is plenty of space:

[root@V-3242 shm]# df -h
Filesystem Size Used Avail Use% Mounted on
/dev/simfs 150G 6.1G 144G 5% /
none 3.9G 776M 3.2G 20% /dev

[root@V-3242 shm]# free -m
total used free shared buffers cached
Mem: 7955 800 7155 0 0 0
-/+ buffers/cache: 800 7155
Swap: 859 773 85

Any ideas? I notice that my Swap is now full, perhaps it could be restricted to only swap memory. If so, how do I change this so it can use my physical RAM?

Here is my fstab if it helps:

none /dev/pts devpts rw 0 0

TobiSGD · 12-20-2011, 10:54 AM

When you want to write to /dev/shm, why do you mount /dev/pts instead?

amunro · 12-20-2011, 11:40 AM

Quote:

Originally Posted by TobiSGD

When you want to write to /dev/shm, why do you mount /dev/pts instead?

I don't, that's just how the fstab came came with the installation. You think I should try writing to /dev/pts ?

TobiSGD · 12-20-2011, 11:45 AM

No, of course not. The best way to use a ramdisk nowadays is doing it with tmpfs. Try something like this:

Code:

mkdir /ramdisk
mount -t tmpfs tmpfs /ramdisk

This will mount the tmpfs ramdisk to /ramdisk and gives it by default a maximal size of 50% of your RAM. Keep in mind that tmpfs only uses the amount of RAM it actually needs. I, for example, use it for /tmp on my laptop with SSD.
This is how the entry in my fstab looks like:

Code:

tmpfs		/tmp		tmpfs		defaults,mode=1777	0 0

Here some output from mount and df:

Code:

tobi@dragon ~ :) % mount
/dev/sda2 on / type ext4 (rw,noatime,discard,commit=0)
proc on /proc type proc (rw)
sysfs on /sys type sysfs (rw)
/dev/sda3 on /home type ext4 (rw,noatime,discard,commit=0)
tmpfs on /dev/shm type tmpfs (rw)
tmpfs on /tmp type tmpfs (rw,mode=1777)
tmpfs on /home/tobi/.cache type tmpfs (rw,mode=1777)
//server/server on /server type cifs (ro,mand)
tobi@dragon ~ :) % df -h
Filesystem       Size  Used Avail Use% Mounted on
/dev/sda2        9.9G  4.3G  5.2G  46% /
/dev/sda3         25G   20G  3.7G  85% /home
tmpfs            1.9G     0  1.9G   0% /dev/shm
tmpfs            1.9G   92K  1.9G   1% /tmp
tmpfs            1.9G  8.0K  1.9G   1% /home/tobi/.cache
//server/server  1.8T  1.8T   30G  99% /server

amunro · 12-20-2011, 11:25 PM

Quote:

Originally Posted by TobiSGD

No, of course not. The best way to use a ramdisk nowadays is doing it with tmpfs. Try something like this:

Code:

mkdir /ramdisk
mount -t tmpfs tmpfs /ramdisk

This will mount the tmpfs ramdisk to /ramdisk and gives it by default a maximal size of 50% of your RAM. Keep in mind that tmpfs only uses the amount of RAM it actually needs. I, for example, use it for /tmp on my laptop with SSD.

Ok I have tried making a new ramdisk however I still find that it copies the data it says 'No space left on device' even though I have plenty of physical RAM left over.

Here is the output I get so you can see what I am talking about:

Code:

 
[root@V-3242 ~]# free -m
             total       used       free     shared    buffers     cached
Mem:          7955       5778       2177          0          0          0
-/+ buffers/cache:       5778       2177
Swap:          859        724        134
[root@V-3242 ~]# df -h
Filesystem            Size  Used Avail Use% Mounted on
/dev/simfs            150G  6.1G  144G   5% /
none                  3.9G  4.0K  3.9G   1% /dev
tmpfs                 3.9G  778M  3.2G  20% /ramdisk
[root@V-3242 ~]# cp server/server.log /ramdisk/server
cp: writing `/ramdisk/server/server.log': No space left on device

As you can see, there is at-least 2177mb left and the /ramdisk directory is only 20% full, however it still says 'No space left on device'!

TobiSGD · 12-21-2011, 10:34 AM

Is it possible that your data contains many small files, so that the file-system runs out of inodes? You can check that with df -i

syg00 · 12-21-2011, 04:40 PM

Why no (real) info ? What is the size of that log file, and what's on /ramdisk (ls /ramdisk) ?

amunro · 12-21-2011, 11:31 PM

No the inodes are fine:

Code:

Filesystem            Inodes   IUsed   IFree IUse% Mounted on
/dev/simfs           119832704  332495 119500209    1% /
none                 1018330      96 1018234    1% /dev
tmpfs                1018330   35182  983148    4% /ramdisk

The folder contains around 35k data files.

Code:

[root@V-3242 ramdisk]# find . -type f | wc -l
35020

Also the size of the file doesn't even matter, it refuses anything I try copy. I can delete everything in the folder however it always reports 'No space left on device' at 20% capacity.

amunro · 12-23-2011, 06:15 AM

Bump.

karanis · 03-06-2012, 12:38 PM

Quote:

Originally Posted by TobiSGD

Is it possible that your data contains many small files, so that the file-system runs out of inodes? You can check that with df -i

TobiSGD is totally right... It is because of being "out of inode", not disk space... At least for my situation.

Thanks TobiSGD