[Gluster-users] Memory overcommit

[Michal Fiala]

Hallo,

we are using glusterfs for 2 years, nowadays version 3.2.7. It works
without problems, but we have a huge memory overcommit. We have system
with 32G ram, 35 glusterfs mounts (35 glusterfs processes),
configuration see bellow. Memory overcommit grows in 1 mount by 32G, see
https://dl.dropbox.com/u/2061501/memory-month.png.
We are using native glusterfs client, fuse fuse-2.8.1, gentoo kernel
3.1.5-hardened. pmap -x output for glusterfs client see
https://dl.dropbox.com/u/2061501/pmap.txt

How can we reduce memory overcommit?

Thanks


One glusterfs server brick configuration

glusterfsd.conf:

volume d1_posix
    type storage/posix
    option directory /data/d1
end-volume

volume d1_locks
    type features/locks
    subvolumes d1_posix
end-volume

volume d1
    type performance/io-threads
    option thread-count 16
    subvolumes d1_locks
end-volume

volume server
    type protocol/server
    option transport-type tcp
    option transport.socket.bind-address glusterfsd
    option volume-filename.d1-key /etc/glusterfs/d1.vol
    option volume-filename.d4-key /etc/glusterfs/d4.vol
    option volume-filename.d7-key /etc/glusterfs/d7.vol
    subvolumes d1 d4 d7
    option auth.addr.d1.allow 192.168.24*
    option auth.addr.d4.allow 192.168.24*
    option auth.addr.d7.allow 192.168.24*
end-volume

d1.conf:

volume client
    type protocol/client
    option transport-type tcp
    option remote-host rajsta1san
    option remote-subvolume d1
end-volume

volume iot
    type performance/io-threads
    option thread-count 8
    subvolumes client
end-volume

Client mounts via fstab record

rajsta1san:d1           /web/data/d1    glusterfs
noauto,noatime,log-file=/var/log/glusterfs/d1.log       0 0