Lustre discuss - Dec 2006 - a network RAID 0+1 with lustre?...

Hi all,

I am currently testing lustre to check if it can possibly replace my NFS 
mounts...

My files are spread on multiple NFS server so that, if I lose 1 NFS 
server I only use access to data located on this server which is solved 
with data written twice. stable enough but not very scalable, isn''t
it?...

Of cours, I expect a lot more from lustre scalability...

I understood that I am able to...
... write to one file system which is hosted on multiple servers
... write to one file system which is hosted by one server and mirror on 
a second one

I''d like to mix both with some kind of mirror for each of my server 
hosting a peace of my lustre filesystem. This way, I would be very 
scalable and still fault tolerant.

Is this possible? Thanks for your help!

Regards,
Nicolas.

Nicolas,

from what I understood you want to write the file once and have it written
twice (on two different OSS) servers. Lustre calls it as ''Multi-server
file
RAID-1''. According with Lustre''s roadmap (
http://www.clusterfs.com/lustre-roadmap-2006.pdf) this feature I''ll be
present at the 1.8 version of Lustre.

Nowadays you already use the best high availability resource I know for
Lustre (every OSS has it''s mirror).

Best regards,
Daniel Colchete

On 12/29/06, Nicolas Bogucki <nbogucki@ina.fr>
wrote:
>
> Hi all,
>
> I am currently testing lustre to check if it can possibly replace my NFS
> mounts...
>
> My files are spread on multiple NFS server so that, if I lose 1 NFS
> server I only use access to data located on this server which is solved
> with data written twice. stable enough but not very scalable,
isn''t it?...
>
> Of cours, I expect a lot more from lustre scalability...
>
> I understood that I am able to...
> ... write to one file system which is hosted on multiple servers
> ... write to one file system which is hosted by one server and mirror on
> a second one
>
> I''d like to mix both with some kind of mirror for each of my
server
> hosting a peace of my lustre filesystem. This way, I would be very
> scalable and still fault tolerant.
>
> Is this possible? Thanks for your help!
>
> Regards,
> Nicolas.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://mail.clusterfs.com/pipermail/lustre-discuss/attachments/20061229/e5bcf351/attachment.html

On 12/29/06, Nicolas Bogucki <nbogucki@ina.fr> wrote:
> Date: Fri, 29 Dec 2006 15:19:33 +0100
> From: Nicolas Bogucki <nbogucki@ina.fr>
> Subject: [Lustre-discuss] a network RAID 0+1 with lustre?...
> To: lustre-discuss@clusterfs.com
> Message-ID: <459523F5.5060704@ina.fr>
> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
> 
> Hi all,
> 
> I am currently testing lustre to check if it can possibly replace my NFS
mounts...
> 
> My files are spread on multiple NFS server so that, if I lose 1 NFS server
I only use access to data located on this server > which is solved with data
written twice. stable enough but not very scalable, isn''t it?...
> 
> Of cours, I expect a lot more from lustre scalability...
> 
> I understood that I am able to...
> ... write to one file system which is hosted on multiple servers
> ... write to one file system which is hosted by one server and mirror on a
second one
> 
> I''d like to mix both with some kind of mirror for each of my
server hosting a peace of my lustre filesystem. This way, I would be very
scalable and still fault tolerant.
> 
> Is this possible? Thanks for your help!
> 
> Regards,
> Nicolas.
 
Hi Nicolas,
 
I believe that the scalability and fault tolerance you require can be achieved
with current Lustre technology plus Linux clustering software of some sort.
Define a Lustre filesystem that has two OSTs, one on each of a pair of servers.
Configure the clustering software to make each server take over serving the
other''s OST if the other goes down. That way you (a) split the load
over both servers when they''re both up and (b) have access to the full
filesystem as long as either server is up. This sort of approach is used
successfully in HP''s SFS (Scalable File Share) product.

>  
> Hi Nicolas,
Hi!
>  
> I believe that the scalability and fault tolerance you require can be
achieved with current Lustre technology plus Linux clustering software of some
sort. Define a Lustre filesystem that has two OSTs, one on each of a pair of
servers. Configure the clustering software to make each server take over serving
the other''s OST if the other goes down. That way you (a) split the load
over both servers when they''re both up and (b) have access to the full
filesystem as long as either server is up. This sort of approach is used
successfully in HP''s SFS (Scalable File Share) product.
This mean that each serveur of my distributed RAID 0 Filesystem will be 
in fact a 2 node cluster, is that what you mean?

Sounds nice! I''ll try this and let you know.

Thanks
Nicolas.