[Gluster-devel] Full bore.

Anand Avati avati at zresearch.com
Fri Nov 16 15:55:19 UTC 2007


Chris,
 what is the client config? only the server config is attached.. the
read-ahead is not really tuned well for gig/e. try smaller page-count (like
2) with a page size of 128kB. You could even try without read-ahead.

2007/11/16, Chris Johnson <johnson at nmr.mgh.harvard.edu>:
>
> On Fri, 16 Nov 2007, Anand Avati wrote:
>
>       Very simple iozone call
>
>       iozone -aN -r 32k -s 131072k
>
> Now what I probably should be doing is testing this against the actual
> load, the real image processing code to be used.  Did look at io
> caching.  That would possibly help if a number of jobs on the cluster
> were to crunch the same file at the same time.  If that were not the
> case I'm not sure what io caching would give me.
>
>       The config?  You can't get much simpler than this, maybe by
> pulling the read-ahead and write-behind which weren't there
> originally.
>
> volume brick1
>    type storage/posix
>    option directory /home/sdm1
> end-volume
>
> volume server
>    type protocol/server
>    subvolumes brick1
>    option transport-type tcp/server     # For TCP/IP transport
> #  option client-volume-filename /etc/glusterfs/glusterfs-client.vol
>    option auth.ip.brick1.allow *
> end-volume
>
> volume writebehind
>    type performance/write-behind
>    option aggregate-size 131072 # in bytes
>    subvolumes brick1
> end-volume
>
> volume readahead
>    type performance/read-ahead
>    option page-size 65536 ### in bytes
>    option page-count 16 ### memory cache size is page-count x page-size
> per file
>    subvolumes brick1
> end-volume
>
> >>
> >>       Performance is still way below NFS.  I have turned on
> >> write-behind and read-ahead.  And with fuse it's a little faster. but
> >> not significantly.
> >
> >
> > What is your IO pattern? what are the tests you are running? what is
> your
> > interconnect? NFS can provide really good re-read performance which you
> can
> > get out of glusterfs using io-cache on the client side.
> >
> >      Can we have a discussion on whether I'm heading in the right
> >> direction and what order things go in for the config files?
> >
> >
> > Can you paste your config files on pastebin and link it and also tell us
> the
> > kind of application/IOtests you are running?
> >
> > thanks,
> > avati
> >
> >
> > --
> > It always takes longer than you expect, even when you take into account
> > Hofstadter's Law.
> >
> > -- Hofstadter's Law
> >
>
>
> -------------------------------------------------------------------------------
> Chris Johnson               |Internet: johnson at nmr.mgh.harvard.edu
> Systems Administrator       |Web:
> http://www.nmr.mgh.harvard.edu/~johnson
> NMR Center                  |Voice:    617.726.0949
> Mass. General Hospital      |FAX:      617.726.7422
> 149 (2301) 13th Street      |Do not meddle in the affairs of wizards, for
> Charlestown, MA., 02129 USA |you are crunchy and good with ketchup.
>
> -------------------------------------------------------------------------------
>



-- 
It always takes longer than you expect, even when you take into account
Hofstadter's Law.

-- Hofstadter's Law



More information about the Gluster-devel mailing list