Gluster users - Feb 2010 - GlusterFS and automated balancing

Hi,

for my small webhosting (3 servers, more to come hopefully) I am 
investigating cluster filesystems. I have seen a few now and I love the 
flexibility that GlusterFS brings. Still I cannot see a way to adapt it 
to suit my needs. I have the following hardware:
- Server #1 with 160GB S-ATA
- Server #2 with 2x 400GB S-ATA
- Server #3 with 2x 1,5TB S-ATA

I am hoping to find a filesystem that fulfills the following requirements:
1. POSIX compliant (Apache, Postfix, etc. will use it) - GlusterFS has it
2. combine the harddisks of all servers into one single filesystem - 
DHT/unify seem to do the job
3. redundancy: have a copy of each single file on at least 2 machines 
such that a single host may fail without people noticing - looks like 
this may be achieved by having AFR below DHT/Unify
4. after a server failure redundancy should automatically be recreated 
(ie. create new copies of all files that only exist once after the crash)
5. just throw in new hardware, connect it with the cluster and let the 
filesystem take care of filling it with data

Hadoop seems strong on points 2.-5. but fails in 1. and is unsuited for 
small files. For GlusterFS however, I cannot see how to achieve 4.-5. 
There always seems to be manual reconfiguration and data movement 
involved, is this correct? Since most of the Wiki is still based on 2.0 
and there is 3.0 out now, I may be missing something.

Hoping for your comments.

Thanks and regards,
Eros

Eros Candelaresi wrote:
> 4. after a server failure redundancy should automatically be recreated 
> (ie. create new copies of all files that only exist once after the crash)
This is done by replicate. When a server comes back up the data is 
"self-healed", ensuring the copies are in
sync.
> 5. just throw in new hardware, connect it with the cluster and let the 
> filesystem take care of filling it with data
This is possible by adding subvolumes to distribute. Currently this 
requires a remount on all clients. With the 3.1 release you'll be able 
to add servers without a remount.

Vikas

H
i all,

I'm running Gluster 3.0.0 on top of XFS and while running a FORTRAN code
that works perfectly well on any other file system, I get runtime errors
when trying to open files -- along the lines of:

At line 386 of file main.f (unit = 18, file = '')
Fortran runtime error: File 'CHGCAR' already exists

Are there known issues with FORTRAN I/O and Gluster?  Is this some sort
of caching artifact?  Its not a consistent problem as it only seems to
happen when running jobs within my scheduling environment (I use SGE).

Let me know if you need more info.

Thanks in advance,
-Brian

-- 
Brian Smith
Senior Systems Administrator
IT Research Computing, University of South Florida
4202 E. Fowler Ave. ENB204
Office Phone: +1 813 974-1467
Organization URL: http://rc.usf.edu


On Thu, 2010-02-11 at 15:13 +0100, Eros Candelaresi
wrote:
> Hi,
> 
> for my small webhosting (3 servers, more to come hopefully) I am 
> investigating cluster filesystems. I have seen a few now and I love the 
> flexibility that GlusterFS brings. Still I cannot see a way to adapt it 
> to suit my needs. I have the following hardware:
> - Server #1 with 160GB S-ATA
> - Server #2 with 2x 400GB S-ATA
> - Server #3 with 2x 1,5TB S-ATA
> 
> I am hoping to find a filesystem that fulfills the following requirements:
> 1. POSIX compliant (Apache, Postfix, etc. will use it) - GlusterFS has it
> 2. combine the harddisks of all servers into one single filesystem - 
> DHT/unify seem to do the job
> 3. redundancy: have a copy of each single file on at least 2 machines 
> such that a single host may fail without people noticing - looks like 
> this may be achieved by having AFR below DHT/Unify
> 4. after a server failure redundancy should automatically be recreated 
> (ie. create new copies of all files that only exist once after the crash)
> 5. just throw in new hardware, connect it with the cluster and let the 
> filesystem take care of filling it with data
> 
> Hadoop seems strong on points 2.-5. but fails in 1. and is unsuited for 
> small files. For GlusterFS however, I cannot see how to achieve 4.-5. 
> There always seems to be manual reconfiguration and data movement 
> involved, is this correct? Since most of the Wiki is still based on 2.0 
> and there is 3.0 out now, I may be missing something.
> 
> Hoping for your comments.
> 
> Thanks and regards,
> Eros
> 
> 
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://gluster.org/cgi-bin/mailman/listinfo/gluster-users