|
From: | Anand Babu Periasamy |
Subject: | Re: [Gluster-devel] How namespace works in mainline-2.5? |
Date: | Fri, 22 Jun 2007 03:12:10 -0700 |
Dale, your assumptions are correct. Here is more info.. PREVIOUS DESIGN: GlusterFS/unify will parallely query every sub-volume (brick) in the cluster. The brick with the file responds with a file descriptor successfully. This model is fast enough and simple. NEW DESIGN: GlusterFS/unify uses a cache to lookup the name space (to find where a file is located). Though it takes an extra call before open, it minimizes lookup operations. There are also other reasons why we implemented name space cache and made it mandatory. * GlusterFS can avoid creation of files that already exist on afailed brick (temporarily down). * Rarely some tools use inodes instead of file names. GlusterFS
issues globally unique inodes consistently through the cache. Cache is stored on a regular POSIX volume. It can be local or remote or even over AFR'd remote volume. Having a cache volume remotely is better because, the cache is shared across all the clients. Remember it is just a cache. If you delete this cache directory, it will rebuild itself transparently. -- Anand Babu Periasamy Dale Dude writes:
Im giving input even though Im not positive. The namespace volume is needed because all the files/dirs are created there. Files being 0 bytes. It seems to be used as the lookup "database" for a set of volumes. So the space requirements are low and I dont think its disposable. I believe they will be using this to keep the AFR volumes in sync. Regards, Dale Brent A Nelson wrote:On Thu, 21 Jun 2007, Dale Dude wrote:As for the doc/example...I see the cluster-client.vol was fixed, but the bricks-ns isnt configured in any of the cluster-server#.vol files.Speaking of which, what are the requirements/recommendations for the namespace cache volume? How big should it be? Any special considerations if we're targetting fault-tolerance (e.g., RAID or AFR underneath), or is it truly disposable? Thanks, Brent _______________________________________________ Gluster-devel mailing list address@hidden http://lists.nongnu.org/mailman/listinfo/gluster-devel_______________________________________________ Gluster-devel mailing list address@hidden http://lists.nongnu.org/mailman/listinfo/gluster-devel
-- Anand Babu GPG Key ID: 0x62E15A31 Blog [http://ab.freeshell.org] The GNU Operating System [http://www.gnu.org]
[Prev in Thread] | Current Thread | [Next in Thread] |