[Linux-cluster] optimising DLM speed?

[Steven Whitehouse]

Hi,

On Wed, 2011-02-16 at 19:36 +0000, Alan Brown wrote:
> > A faster way to just grab lock numbers is to grep for gfs2
> in /proc/slabinfo as that will show how many are allocated at any one
> time.
> 
> True, but it doesn't show mow many are used per fs.
> 
For the GFS2 glocks, that doesn't matter - all of the glocks are held in
a single hash table no matter how many filesystems there are. The DLM
however has hash tables for each lockspace (per filesystem) so it might
make a difference there.

> FWIW, here are current stats on each cluster node (it's evening and 
> lightly loaded)
> 
> 
> gfs2_quotad           47    108    144   27    1 : tunables  120   60 
>   8 : slabdata      4      4      0
> gfs2_rgrpd          9563   9618    184   21    1 : tunables  120   60 
>   8 : slabdata    458    458      0
> gfs2_bufdata      318804 318840     96   40    1 : tunables  120   60 
>   8 : slabdata   7971   7971      1
> gfs2_inode        725605 725605    800    5    1 : tunables   54   27 
>   8 : slabdata 145121 145121      0
> gfs2_glock        738297 738297    424    9    1 : tunables   54   27 
>   8 : slabdata  82033  82033      0
> 
> gfs2_quotad           94    189    144   27    1 : tunables  120   60 
>   8 : slabdata      7      7      0
> gfs2_rgrpd          1658   1680    184   21    1 : tunables  120   60 
>   8 : slabdata     80     80      0
> gfs2_bufdata      1065806 1067080     96   40    1 : tunables  120   60 
>     8 : slabdata  26677  26677      0
> gfs2_inode        986986 1024845    800    5    1 : tunables   54   27 
>    8 : slabdata 204969 204969      0
> gfs2_glock        1105575 1812825    424    9    1 : tunables   54   27 
>     8 : slabdata 201425 201425      1
> 
> gfs2_quotad           45    108    144   27    1 : tunables  120   60 
>   8 : slabdata      4      4      2
> gfs2_rgrpd          6515   6573    184   21    1 : tunables  120   60 
>   8 : slabdata    313    313      0
> gfs2_bufdata      100785 101000     96   40    1 : tunables  120   60 
>   8 : slabdata   2525   2525      0
> gfs2_inode        2954515 2954515    800    5    1 : tunables   54   27 
>     8 : slabdata 590903 590903      0
> gfs2_glock        3332311 3639843    424    9    1 : tunables   54   27 
>     8 : slabdata 404427 404427      0
> 
Thanks for the info. There is now a bug open (bz #678102) for increasing
the default DLM hash table size,

Steve.

> 
> --
> Linux-cluster mailing list
> Linux-cluster at redhat.com
> https://www.redhat.com/mailman/listinfo/linux-cluster