On Jan 22, 2007, at 1:44 PM, [EMAIL PROTECTED] wrote:
>
> On Jan 22, 2007, at 12:32 PM, [EMAIL PROTECTED] wrote:
>
>> Hi.
>> now i use pvfs in the university and all the i/o nodes are in the
>> same network (ee network).
>> If i want to add i/o nodes from other networks in the university
>> how ca do i solve the login problems?
> Hi Jakob,
>
> To clarify, you are using PVFS version 1.x, correct? Are there
> limitations you have that force you to stick with version 1.x,
> instead of using to 2.6.2 (such as a lot of data stored in the old
> 1.x format)? Otherwise, can I encourage you to upgrade to PVFS
> version 2? Its more actively being maintained (this list is
> especially for PVFS version 2.x questions only).
>
> As for login problems, can you be more specific about the setup you
> want to achieve? I'm not that familiar with PVFS version 1.x, but in
> version 2.x, the servers can run as any user, so if the user
accounts
> you have on the two networks differ, you should still be able to run
> the servers on them just fine.
>
> Perhaps Walt can answer your question in a PVFS 1.x context if you
> provide more info about the problems you're having.
>
> -sam
>> thanks
>> Jakob
>>
>> _______________________________________________
>> Pvfs2-users mailing list
>> [email protected]
>> http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
>
Thank u.
Yes, i am using pvfs 1.6.3 but its temporary i just want to check
the performance of
the pvfs versus the nfs.
And after i will consider to upgrade to pvfs 2.
Anyway my question was for pvfs 2.
So i understand that its possible even if i have 2 differ accounts.
how can it be?
The PVFS v2 server manages the uid/gids itself, independent of the
accounts on server (metadata or IO) node it runs on. The PVFS v2
client uses the uid/gids of the host it runs on.
do i have to insert my user and password to some file?
No. Once you mount a new pvfs volume on the client, that mountpoint
acts as a scratch space (like /tmp). As you create files and
directories within that mountpoint, the uid/gid on that host (that
has the pvfs volume mounted) are used (stored on and accessed from
the metadata server). This means that all clients that mount the
same pvfs volume should have the same account information (using NIS
for example), otherwise you'll get into some odd cases where files
created on one client don't have matching uid/gid entries on another.
If you really want to do what you asked in your previous email, i.e.
setup multiple pvfs _servers_ across some sort of network, and access
them all from the same client (or set of clients within the same
network), the only restriction you have is that the ports you use on
the servers within one (or both) of the networks aren't firewalled
off from the clients.
Remember though that your performance is likely going to be much
worse going over a wide area network than local, and since files are
striped over all the IO servers in pvfs, your performance will be
limited by the slowest link (over the WAN). So while its possible to
do what you want, its probably not a good idea in most cases.
If you're interested in exploring how data is distributed over pvfs
servers based on server location, there are other distribution models
besides the default pvfs round-robin striping, and research is being
done with pvfs by different groups in this area.
another thing how can i get programs that will proof the
performance of the pvfs.
especially in reading from big text file.
Once the pvfs volume has been mounted, you can use any of the many
filesystem benchmarks floating around, or just use dd. If you don't
want to load the pvfs module and mount the volume, you can use the
pvfs2-cp app, which gets installed with pvfs by default. It has a -t
option that provides performance and timing info of the copy.
-sam
thank u
Jakob.
_______________________________________________
Pvfs2-users mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
