On Apr 23, 2007, at 5:17 PM, David Brown wrote:
Naturally, let us know (good or bad) how PVFS does with such a large
number of servers and clients. We have some ideas for file layouts
that might make sense for 1k parallel clients.
I'll be sure to do that, I come from the open source gpl'ed world.
Its very nice to know when all that time and effort you put into
making a really good piece of software either blows up or works like a
charm :)
The server can run as a non-root user. The pvfs2-client needs to be
able to read the /dev/pvfs2-req device. We always run pvfs2-
client as
root, but it might be possible to run as some other user once you
change the device permissions apropriately.
Oooo, installing a nice udev rules file might be appropriate for
setting permissions and such for this.
As I see it, if you need root for one step, then you might as well
use it for both. The kernel module has to be loaded as root anyway,
so what does running the client as a non-root user buy you?
However, it seems the kernel
module doesn't actually try and create the device... pvfs2-client
creates the device... is there any reason why the kernel module for
2.6 doesn't try to request that the device be created?
I think its probably because its just easier to use mknod then figure
out how to use the syscall table in the kernel, but if you know of a
better way, we'd certainly consider any patches you send. In
general, the pvfs model is keep the kernel module as lightweight as
possible, so anything that can go in the userspace client gets put
there.
-sam
- David Brown
_______________________________________________
Pvfs2-developers mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers
_______________________________________________
Pvfs2-developers mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers
