[Gluster-devel] more details about "No metadata dependencies" ??

[Yi Ling]

2009/7/31 Raghavendra G <raghavendra at gluster.com>

> 2009/7/31 Yi Ling <lingyi.pro at gmail.com>:
> >
> >
> > 2009/7/31 Harald Stürzebecher <haralds at cs.tu-berlin.de>
> >>
> >> Hello!
> >>
> >> 2009/7/31 Yi Ling <lingyi.pro at gmail.com>:
> >> > there is a section about metadata on page
> >> > http://www.gluster.com/news.php
> >> >
> >> > "No metadata dependencies - metadata management is frequently a source
> >> > of
> >> > performance bottlenecks, particularly when there are a very large
> number
> >> > of
> >> > files, and can be a single point of failure in traditional file
> systems.
> >> > GlusterFS eliminates this problem with intelligent data placement, an
> >> > automatic data distribution method that optimizes the use of resources
> >> > and
> >> > eliminates metadata bottlenecks."
> >> >
> >> > yes,it's a advantage of GlusterFS. but yet i found no more detailed
> >> > information about the "intelligent data placement" "an automatic data
> >> > distribution methed". could anyone explain it in detail? and something
> >> > about
> >> > file managemment~~~
> >>
> >> AFAIK, the GlusterFS metadata is stored as extended attributes to the
> >> files in the underlying filesystem.
> >>
> >> Hints can be found here:
> >>
> >>
> >>
> http://www.gluster.org/docs/index.php/Understanding_AFR_Translator#Locking_and_Change_Log
> >>
> >>
> >>
> http://www.gluster.org/docs/index.php/Setting_up_AFR_on_two_servers_with_pre-existing_data#Preparing_the_directories
> >>
> >>
> >> Harald Stürzebecher
> >
> > thanks for replying~~ but im still confused~~~
> >
> > in client, what's the routine of an operation like "create"/"read" ,from
> > client to server (even backend filesystem)?
> >
> > "the GlusterFS metadata is stored as extended attributes to the files in
> the
> > underlying filesystem"
> > well, after storing, how could i read the file and how to know where the
> > file is located (or which server/brick the file is stored)
>
> Distribute Translator provides the functionality of aggregating the
> storage from individual nodes and giving the unified view of the total
> storage. It stores the layout (which says, which file goes into which
> node) of the directory in extended attributes of the directory. And
> directories are created on all the nodes. Files are created on only
> one of the nodes.
>
> >
> >
> > _______________________________________________
> > Gluster-devel mailing list
> > Gluster-devel at nongnu.org
> > http://lists.nongnu.org/mailman/listinfo/gluster-devel
> >y
> >
>
>
> regards,
> --
> Raghavendra G
>

yes,  HDT translator depends on its name hash, and unify translator has
separate namespace volume which is shared by all clients.   as i know,
"currently each translator has its own mechanism to maintain global
namespace."  but except from hdt and unify, how do the other tanslators
implement the same operation (create, read,readdir ...) ...
afr,map,ha,stripe etc.

now i have no common concept about the mechanism, so im so confused to
understand the routine of different translator.  thanks for your patience
very much ~~~
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://supercolony.gluster.org/pipermail/gluster-devel/attachments/20090731/1cfd07dd/attachment-0003.html>