Lustre discuss - Jan 2007 - Network Raid 5 setup

Hi,


I am looking for a network storage filesystem for our network. While
browsing the web i came across lustre but i have a few questions so i
hope this is the right place to ask them.


1. Can i use lustre like a network raid 5 storage system if one of the
nodes fail do the others have a copy of the complete data to continue
working.

2. Can i add further servers to expand the cluster when needed. And do
the disk sizes need to be identical.

3. Can these servers have normal disks or do i need shared storage to
run lustre.

4. How can i access my lustre filesystem from a windows os.

5. What are the common ways of accessing or mounting a lustre
filesystem.

6. Which distro would you recommend to run lustre on. At the moment i am
running gentoo on most of my servers which would mean patching a
specific kernel if i understand that right.


My idea was to start of with three servers with about 500 GB of disk
space each. And combine this servers to one storage filesytem which has
to be redundant to prevent data loss. I need to be able to access this
filesystem from windows and linux servers. I would also like to be able
to extend this setup to 5 or more servers when the storage space is
needed. Can lustre meet this requirements? Or do i need to look further
for an other solution. I would be grateful for any help or advice
someone on the mailing list may be able to offer.


Thanks Regards


Simon

I''ll try and answer below.  Disclaimer: I am not a CFS guy...   
> 
> 1. Can i use lustre like a network raid 5 storage system if 
> one of the nodes fail do the others have a copy of the 
> complete data to continue working.
Raid5 mode is not avilible in lustre today, there is a possibility in
the future.  Mostly people run raid underneath lustre.  Also there is a
possibility for failover between two hosts using a shared disk.
> 
> 2. Can i add further servers to expand the cluster when 
> needed. And do the disk sizes need to be identical.
Yes expansion is very possible, and getting easier with each release.
The latest beta''s even allow removal of devices.  The disk sizes do not
need to be identical.

> 3. Can these servers have normal disks or do i need shared 
> storage to run lustre.
Any block device.  Shared storage is nice when you want to set up
failover pairs.
> 4. How can i access my lustre filesystem from a windows os.
Samba/CIFS exports, which can run a little slower, also scp, ftp, work
as well.  At tone point CFS was talking about a natic windows client,
but I havent heard much lately.
> 5. What are the common ways of accessing or mounting a lustre 
> filesystem.
Mounting it directly on linux nodes.  We share out scp, ftp, and samba
to our archive users.
> 6. Which distro would you recommend to run lustre on. At the 
> moment i am running gentoo on most of my servers which would 
> mean patching a specific kernel if i understand that right.
The CFS team directly supports the Redhat series of rpm''s and you can
download RPMS directly for that. Suse rpms are also availible.  We roll
our own kernels most of the time, here at PNNL.

> My idea was to start of with three servers with about 500 GB 
> of disk space each. And combine this servers to one storage 
> filesytem which has to be redundant to prevent data loss. I 
> need to be able to access this filesystem from windows and 
> linux servers. I would also like to be able to extend this 
> setup to 5 or more servers when the storage space is needed. 
> Can lustre meet this requirements? Or do i need to look 
> further for an other solution. I would be grateful for any 
> help or advice someone on the mailing list may be able to offer.
The only issue I see here is not lustre provided redundant data.  You
Would need to provide redundancy on the servers underneath lustre.


Evan Felix
PNNL