[Gluster-devel] AFR between two bricks over 3000 miles

Anand Avati avati at zresearch.com
Mon Mar 3 11:09:43 UTC 2008


the bottleneck seems to be create/open which is synchronous over all
subvolumes. do you have numbers without involving the remote site/afr ?

avati

2008/3/1, nathan at robotics.net <nathan at robotics.net>:
>
> On Sat, 1 Mar 2008, Krishna Srinivas wrote:
>
> > Nathan,
> >
> > Check the performance by bringing writebehind into the picture your
> > config file shows that WB and iocache translators are not being used.
>
>
> Sorry, that was stupid, changed it to subvolumes nyc sjc_write-behind that
> changed the delay to real    0m53.005s, still a far cry from .03 I get
> local. This is only 748K, so I would expect it to write local and write
> behind the 800K to the remote site.
>
>
> >
> > On Sat, Mar 1, 2008 at 8:48 AM,  <nathan at robotics.net> wrote:
> >>
> >>  /etc/sysconfig is only 748K
> >>  cp /etc/sysconfig to /share/mirror took real    0m55.319s
> >>  cp /etc/sysconfig to /share took real    0m0.030s
> >>
> >>  I expected it to be much faster. :)
> >>
> >>  # Client
> >>  volume nyc
> >>    type protocol/client
> >>    option transport-type tcp/client
> >>    option remote-host 10.11.0.1
> >>    option remote-subvolume nyc
> >>  end-volume
> >>
> >>  volume sjc
> >>    type protocol/client
> >>    option transport-type tcp/client
> >>    option remote-host 10.12.0.1
> >>    option remote-subvolume sjc
> >>  end-volume
> >>
> >>  volume sjc_iocache
> >>    type performance/io-cache
> >>    option page-size 256KB
> >>    option page-count 2
> >>    subvolumes sjc
> >>  end-volume
> >>
> >>  volume sjc_write-behind
> >>    type performance/write-behind
> >>    option aggregate-size 1MB
> >>    option flush-behind on
> >>    subvolumes sjc_iocache
> >>  end-volume
> >>
> >>  volume mirror
> >>    type cluster/afr
> >>    subvolumes nyc sjc
> >>  end-volume
> >>
> >>
> >>
> >> ><>
> >>  Nathan Stratton
> >>  nathan at robotics.net
> >>  http://www.robotics.net
> >>
> >>
> >>
> >>  On Thu, 28 Feb 2008, Anand Avati wrote:
> >>
> >> > on the client. you also might want to put this write-behind +
> io-cache pair
> >> > in the subvolume path of afr which leads towards the remote site
> alone. Also
> >> > make sure afr does not have the remote site as the first subvolume,
> and has
> >> > the option read-subvolume <local-volume> so that reads are not
> scheduled to
> >> > the remote site.
> >> >
> >> > avati
> >> >
> >> > 2008/2/28, nathan at robotics.net <nathan at robotics.net>:
> >> >>
> >> >>
> >> >> On Thu, 28 Feb 2008, Anand Avati wrote:
> >> >>
> >> >>> using a combination of write-behind with io-threads (with option
> >> >>> flush-behind on) prevents the wait of upto N-MB of data (where N is
> >> >> 'option
> >> >>> cache-size NMB' of io-threads)
> >> >>
> >> >>
> >> >> Client and server are on each host, should I do this on the client
> or
> >> >> server?
> >> >>
> >> >> -Nathan
> >> >>
> >> >
> >> >
> >> >
> >> > --
> >> > If I traveled to the end of the rainbow
> >> > As Dame Fortune did intend,
> >> > Murphy would be there to tell me
> >> > The pot's at the other end.
> >> >
> >>
> >>
> >>  _______________________________________________
> >>  Gluster-devel mailing list
> >>  Gluster-devel at nongnu.org
> >>  http://lists.nongnu.org/mailman/listinfo/gluster-devel
> >>
> >
>



-- 
If I traveled to the end of the rainbow
As Dame Fortune did intend,
Murphy would be there to tell me
The pot's at the other end.



More information about the Gluster-devel mailing list