memory fragmentation on nas

Question

I have a qnap nas unit that I was trying to create an additional iscsi target on, and the creation process fails. Im not trying to do anything I haven't done on other identical units, but this one has been running for longer.

ssh in and have a look

dmesg contains amongst many other things:

[3431220.016334] SLAB: Unable to allocate memory on node 0 (gfp=0xd0)
[3431220.016335]   cache: kmalloc-65536, object size: 65536, order: 4
[3431220.016341]   node 0: slabs: 12/12, objs: 12/12, free: 0

Ok we got the cause of the problem.

# cat /proc/meminfo 
MemTotal:        3939760 kB
MemFree:          190480 kB
Buffers:         3212476 kB
Cached:           188868 kB
SwapCached:           48 kB

So all memory allocated to buffers. Hardly surprising on a linux machine that does little other than block IO

# cat /proc/buddyinfo
Node 0, zone      DMA      8      2      1      1      1      1      1      1      2      2      2 
Node 0, zone    DMA32  22614    389      0      1      1      2      3      1      0      0      0 
Node 0, zone   Normal  12150   7683      0      2      5      2      1      1      0      0      0 

ouch. This is showing some severe memory fragmentation, with virtually no blocks greater than 8K available. Is there any way of clearing the memory fragmentation problem (temporarily reducing the memory allocated to buffers or otherwise) short of restarting the machine, a process that is going to be hugely inconvenient? Also, is it likely that adding more memory to a machine used like this is going to help this particular problem, there are obviously other advantages to upgrading the memory.

Answer 1

Try issuing sync; echo 3 > /proc/sys/vm/drop_caches. This will release all cached/buffered memory.

If this does not work, you can try compacting/defragmenting memory issuing the following command: echo 1 > /proc/sys/vm/compact_memory