Lustre discuss - May 2006 - Could ONE Lustre file system act as multiple logical file systems?

Hello--

On Fri, 2004-07-30 at 04:21, Wenguo WEI wrote:
> 
>     Could ONE Lustre file system act as multiple logical file systems?
> 
>     Suppose a data center save all data into ONE lustre file system,
> there are different data from different applications such as company
> internal office data, customs information and public web page etc,
> then different lustre client/different application client can only
> access what they have permission to access. We know this can be done
> be linux file permission, but does lustre have some special function
> act as ´virtual privary file system¡?
If I understand your question, then strictly speaking the answer is no,
not yet.  It is not possible to mount a sub-directory of a Lustre
namespace as the root, which is I think what you are asking for.

This feature is on our roadmap, but I do not know precisely when it will
enter the production code.

Just in case I didn''t understand your question correctly, I will
mention
two other things:

- standard UNIX permissions are 100% supported, and

- it is possible to serve multiple Lustre file systems from a single MDS

I hope that helps--

-Phil

Hello--

On Wed, 2004-08-04 at 00:36, Ananda Bhattacharya wrote:
> 
> How many Lustre clients can a single MDS handle?
In my opinion, the right question is not how many clients, but how many
concurrent operations.  Lustre runs today on clusters as large as 1,300
nodes, and there is no indication that it would not continue to scale
well if more hardware were available.  I expect that we will see a
5,000-node Lustre file system within 12 months.

On properly configured and tuned MDSs, we have given demonstrations of
1,000 clients doing random lookups at a rate of 5,000/sec, and similar
numbers for random creations (even in a single directory).

With hardware which is not completely commodity, like some nvram for the
file system journal, it should be possible to get even more.
> Can you run 2 MDS and both of them export the same backend FS ?
Only for the purpose of failover, in which only one MDS is active at any
given time.  Metadata clustering is being developed, and large parts
already work well, but it will not be production-quality any time soon.

-Phil