[Gluster-devel] cluster/ec: Added the erasure code xlator

[Xavier Hernandez]

Hi Kaushal,

On Tuesday 13 May 2014 20:22:06 Kaushal M wrote:
> The syntax looks good. If you need help with the cli and glusterd changes,
> I'll be happy to help.
Thanks. It will be really appreciated.

I think I've a working modification. Not sure if there's something else that 
have to be modified. I'll push it for review very soon and add you as a 
reviewer.

I also decided to change the volume option 'size' to 'redundancy', and its 
format. It seems more intuitive now. The 'size' option had the format 'N:R', 
where N was the total number of subvolumes and R the redundancy. Since the 
number of subvolumes can be directly calculated from the 'subvolumes' keyword, 
only the redundancy is really needed.

Xavi

> 
> On Tue, May 13, 2014 at 8:08 PM, Xavier Hernandez 
<xhernandez at datalab.es>wrote:
> > I'm trying to modify the cli to allow the creation of dispersed volumes.
> > 
> > Current syntax for volume creation is like this:
> >     volume create <volname> [stripe <count>] \
> >     
> >                             [replica <count>] \
> >                             [transport <tcp|rdma|tcp,rdma>] \
> >                             <brick>?<vg_name>... \
> >                             [force]
> > 
> > I propose to use this modified syntax:
> >     volume create <volname> [stripe <count>] \
> >     
> >                             [replica <count>] \
> >                             [disperse <count>] \
> >                             [redundancy <count>] \
> >                             [transport <tcp|rdma|tcp,rdma>] \
> >                             <brick>?<vg_name>... \
> >                             [force]
> > 
> > If 'disperse' is specified and 'redundancy' is not, 1 is assumed for
> > redundancy.
> > 
> > If 'redundancy' is specified and 'disperse' is not, disperse count is
> > taken
> > from the number of bricks.
> > 
> > If 'disperse' is specified and the number of bricks is greater than the
> > number
> > indicated (and it is a multiple), a distributed-dispersed volume is
> > created.
> > 
> > 'disperse' and 'redundancy' cannot be combined with 'stripe' or 'replica'.
> > 
> > Would this syntax be ok ?
> > 
> > Xavi
> > 
> > On Tuesday 13 May 2014 12:29:34 Xavier Hernandez wrote:
> > > I forgot to say that performance is not good, however there are some
> > > optimizations not yet incorporated that may improve it. They will be
> > 
> > added
> > 
> > > in following patches.
> > > 
> > > Xavi
> > > 
> > > On Tuesday 13 May 2014 12:23:15 Xavier Hernandez wrote:
> > > > Hello,
> > > > 
> > > > I've just added the cluster/ec translator for review [1].
> > > > 
> > > > It's a rewrite that does not use any additional translator or library.
> > > > It's
> > > > still a work in progress with some bugs to solve, but its architecture
> > > > should be stable. The main missing feature is self-heal, that will be
> > > > added
> > > > once the main code is stabilized and reviewed.
> > > > 
> > > > Feel free to review it and send any comment you think appropiate.
> > > > 
> > > > Thanks,
> > > > 
> > > > Xavi
> > > > 
> > > > [1] http://review.gluster.org/7749
> > > > _______________________________________________
> > > > Gluster-devel mailing list
> > > > Gluster-devel at gluster.org
> > > > http://supercolony.gluster.org/mailman/listinfo/gluster-devel
> > > 
> > > _______________________________________________
> > > Gluster-devel mailing list
> > > Gluster-devel at gluster.org
> > > http://supercolony.gluster.org/mailman/listinfo/gluster-devel
> > 
> > _______________________________________________
> > Gluster-devel mailing list
> > Gluster-devel at gluster.org
> > http://supercolony.gluster.org/mailman/listinfo/gluster-devel