Lustre discuss - May 2007 - Lustre on the OSS?

Hi,

I have a few brief questions on the Lustre OSS architecture (sorry if
these are redundant, I looked but didn''t see these in my Lustre
training
info).

- Do the Lustre modules on the OSS all run entirely in kernel mode?  Is
there anything in the IO path that invokes a user/kernel transition, or
perform any data block copying?

- Where do the Lustre modules hook into the back-end IO stack on the
OSS?  (VFS/VOP?  Block?  SCSI?  submit_bio()?)   I''m interested in
optimizing the Linux IO stack but I don''t want to collide with any
Lustre changes.

- Does Lustre on the OSS make use of the Linux page cache or direct IO?
Is any of that configurable?

- What Linux kernel goes with Lustre 1.6?


Thanks in advance,
Ed Nadolski
LSI Logic

On May 08, 2007  09:32 -0600, Nadolski, Ed wrote:
> - Do the Lustre modules on the OSS all run entirely in kernel mode?
Yes.
> Is there anything in the IO path that invokes a user/kernel transition, or
> perform any data block copying?
For regular userspace IO there is always a copy between userspace and kernel.
If you use TCP then there is another copy on receive (either on the client
or server depending on whether you are reading or writing).
> - Where do the Lustre modules hook into the back-end IO stack on the
> OSS?  (VFS/VOP?  Block?  SCSI?  submit_bio()?)   I''m interested in
> optimizing the Linux IO stack but I don''t want to collide with any
> Lustre changes.
The majority of filesystem operations are done at the VFS/VOP level.
For bulk data IO (read/write) they use submit_bio() (see filter_io_26.c).
> - Does Lustre on the OSS make use of the Linux page cache or direct IO?
> Is any of that configurable?
Not currently.
> - What Linux kernel goes with Lustre 1.6?
Depends which distro you are using.

Cheers, Andreas
--
Andreas Dilger
Principal Software Engineer
Cluster File Systems, Inc.

Along the same lines - Are there plans for ClusterFS to support newer
versions of the vanilla kernel? Current patches exist for  2.6.8 and
2.6.9. I ported them to 2.6.18 and got them to work. We''d like to move
to 2.6.20 (or better) and have official Lustre patches for that version.
Thanks,
Timothy Snider 
Storage Architect
Strategic Planning, Technology and Architecture
LSI Logic Corporation
3718 North Rock Road
Wichita, KS 67226
(316) 636-8736 
tim.snider@lsi.com 


-----Original Message-----
From: lustre-discuss-bounces@clusterfs.com
[mailto:lustre-discuss-bounces@clusterfs.com] On Behalf Of Andreas
Dilger
Sent: Tuesday, May 08, 2007 1:24 PM
To: Nadolski, Ed
Cc: lustre-discuss@clusterfs.com
Subject: Re: [Lustre-discuss] Lustre on the OSS?

On May 08, 2007  09:32 -0600, Nadolski, Ed wrote:
> - Do the Lustre modules on the OSS all run entirely in kernel mode?
Yes.
> Is there anything in the IO path that invokes a user/kernel 
> transition, or perform any data block copying?
For regular userspace IO there is always a copy between userspace and
kernel.
If you use TCP then there is another copy on receive (either on the
client or server depending on whether you are reading or writing).
> - Where do the Lustre modules hook into the back-end IO stack on the
> OSS?  (VFS/VOP?  Block?  SCSI?  submit_bio()?)   I''m interested in
> optimizing the Linux IO stack but I don''t want to collide with any
> Lustre changes.
The majority of filesystem operations are done at the VFS/VOP level.
For bulk data IO (read/write) they use submit_bio() (see
filter_io_26.c).
> - Does Lustre on the OSS make use of the Linux page cache or direct
IO?
> Is any of that configurable?
Not currently.
> - What Linux kernel goes with Lustre 1.6?
Depends which distro you are using.

Cheers, Andreas
--
Andreas Dilger
Principal Software Engineer
Cluster File Systems, Inc.

_______________________________________________
Lustre-discuss mailing list
Lustre-discuss@clusterfs.com
https://mail.clusterfs.com/mailman/listinfo/lustre-discuss