"David M. Grimes" wrote:
>
> Do you know if there is any intent to make the SGI SCSI patches available
> for 2.4 kernels? From the documentation provided with KAIO, it appears that
> when using kaio on files on a ext2 filesystem you get asynchrony independant
> of the slave threads. Is this true? If not, how can I achieve true
> asynchronous I/O without being limited by the number of slave threads?
Lots of questions, let me answer one by one ;-)
1. The 2.2 SGI Raw I/O patches are not directly available for 2.4 kernels.
This is because we'd rather go along with what's in 2.4 & enhance it.
To this end, there are some enhancements available at:
http://oss.sgi.com/projects/rawio/
See the 5/23/00 update. Unfortunately, that patch is now outdated. There
was an updated version available in the XFS release, but I'm not sure;
the person working on it is on holiday leave for about 2 weeks.
2. With ext2 you can get thread-independent asynchrony for reads, yes. Writes
still require slave thread context.
>
> If the SGI SCSI pathches are not (and cannot be) availble for 2.4 kernels,
> can the kaio-1.3 be produced against a 2.2.13 kernel so it can be used with
> the SGI SCSI patches?
I really have limited time & the focus is on 2.4; so the right thing
is to get some form of the below (scsi-sgl) layer in 2.4 ...
>
> > to queue the I/O as compared the SGI SCSI patch which
> > directly translates the I/O into a scsi scatter-gather list.
>
> Ultimately, this is what I am trying to do. I guess I could get by with
> large files on an ext2 filesystem, but I'd really prefer to use raw SCSI I/O.
Yes, try to use ext2 & get other issues sorted out. In a few weeks,
we can look into getting a more performant raw i/o patch.
regards,
--------------------------------------------------------------------------
Rajagopal Ananthanarayanan ("ananth")
Member Technical Staff, SGI.
--------------------------------------------------------------------------
|