On Jul 5, 2007, at 11:53 PM, xiangyong ouyang wrote:


I'm running MPI-IO program to test the write/read performance with pvfs2 file system on a linux cluster. Here are some questions I'm not clear with.

1. For a file in pvfs2 file system, from which IO server the pvfs2 begin to stripe file data? In another word, how to find the first IO server that starts to stripe this file stored in pvfs2 file system?

The first server a file is striped over is chosen randomly. A development patch has been posted to the developers list that allows the first server (and in fact the order of all the servers) to be specified to the PVFS_sys_create function. This support doesn't exist in a PVFS released version or MPI-IO at present.


2. How to change the strip_size? I know we can change the PVFS_DIST_SIMPLE_STRIPE_DEFAULT_STRIP_SIZE value in the source header file: include/pvfs2-dist-simple-stripe.h. But that requires a re-compilation whenever I change the value. Is there a global environment variable that I can set to change the strip size?

You *should* be able to specify the strip size in the config file:

<Distribution>
        Name simple_stripe
        Param strip_size
        Value 524288
</Distribution>

This should be specified within the <Filesystem> context.

There's a bug in the 2.6.3 release that doesn't allow this to work though. I've attached a patch that you can apply that will fix the bug.

-sam

Attachment: config-dist.patch
Description: Binary data



thanks!


xiangyong ouyang

_______________________________________________
Pvfs2-users mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users


_______________________________________________
Pvfs2-users mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users

Reply via email to