Lustre discuss - May 2006 - about lustre in an evolving environement

On Sat, 2004-05-15 at 06:22, Christophe Delaere wrote:
> 
> I''m studying the possibility to use lustre on our local cluster
and/or local
> desktop pool.
> 
> Presently, we have O(100) linux machines (from which 24 are part or a 
> dedicated cluster) with each time a local scratch (also nfs-exported). The 
> idea would be to "merge" those scratch spaces to improve
availability.
> 
> I''ve three questions:
> 
> 1. Say I have a running lustre installation. Can I add a new machine, and 
> insert it transparently in the system as a new OST ?
> 
> 2. Now the opposite. I have a running lustre installation, and one of the 
> linux machines used as OST has to be removed. Can it be done transparently
?
> What appens to the data on that particular disk ?
With today''s versions of Lustre, clients and the MDS must be unmounted
and re-mounted with a new configuration.

Code to support transparent OST addition/removal has been mostly
written, and will probably appear in Lustre 1.6.x later this year.

When the OST is removed, the administrator has two options: he can leave
the OST configured but disabled, in which case -EIO will be returned for
all read, write, and unlink operations.  The other option is to remove
the OST entirely, in which case new objects will be automatically
allocated to replace the now-missing objects.  This has the effect of
making those parts of your files sparse until new data is written.
> 3. Is it a way to introduce redundancy into the system ? I mean, can data
be
> duplicated on 2 machines (2 OSTs) so that one can be shut down ? If desktop
> machines are included in a lustre, it may be worst doing it, as the user
can
> shut down his machine even if we told him not to do so.
We are working on the design and development of a RAID1 LOV, to
automatically mirror data on multiple OSTs.  I don''t have a precise
idea
about when this will become part of a supported release.

Thanks--

-Phil

Hi,

I''m studying the possibility to use lustre on our local cluster and/or
local=20
desktop pool.

Presently, we have O(100) linux machines (from which 24 are part or a=20
dedicated cluster) with each time a local scratch (also nfs-exported). The=20
idea would be to "merge" those scratch spaces to improve availability.

I''ve three questions:

1. Say I have a running lustre installation. Can I add a new machine, and=20
insert it transparently in the system as a new OST ?

2. Now the opposite. I have a running lustre installation, and one of the=20
linux machines used as OST has to be removed. Can it be done transparently ?=20
What appens to the data on that particular disk ?

3. Is it a way to introduce redundancy into the system ? I mean, can data be=20
duplicated on 2 machines (2 OSTs) so that one can be shut down ? If desktop=20
machines are included in a lustre, it may be worst doing it, as the user can=20
shut down his machine even if we told him not to do so.

Sorry for those beginner''s questions, and thanks a lot.
Christophe.

--=20
+-----------------------------------------------------------\|/---+
| Christophe DELAERE            office: e253               !o o!  |
| UCL - FYNU                    phone : 32-(0)10-473234    ! i !  |
| chemin du cyclotron, 2        fax   : 32-(0)10-452183     `-''   |
| 1348 Louvain-la-Neuve BELGIUM e-mail: delaere@fynu.ucl.ac.be    |
+-----------------------------------------------------------------+