Ananth Ananthanarayanan wrote:
> For some reason your patch contains extra files with .orig extensions;
> i gues you had some "saved" copies lying around when you ran the diff.
> Due to this, the patch you sent out has 8000+ lines of diff, whereas
> the original kaio-kern-1.0 has only 2000+ lines of diff. I tried
> manually editing out the .orig diffs, but still couldn't find what
> exactly your changes were. Can you please sending a corrected
> version of the patch? Also, if the differences are not big,
> could you also send the diffs (in text) in your reply.
Patch fixed. It just takes into account a very minor change in sched.h,
as well as some line number changes. This one is almost exactly the same
size as yours.
> Also, re: your question about 2.3.X ... this is one of the highest
> priority items on my todo lists. Note that most of the changes
> come in large chunks, so I'm hoping(!) that we can get away with
> porting a significant portion of the code. I'm also happy to
> see you find KAIO to be an improvement over glibc AIO.
> I hope to get such feedback before pushing the patches onto linux-kernel
> mailing list & thence to Linus et.al. If you (or anyone else
> reading this message) has a quantification of KAIO vs. glibc-aio
> w.r.t. performance or posix-compliance (or just difficult bugs),
> that information will greatly influence in getting KAIO into kernel
> proper.
Well, I can safely say that glibc AIO doesn't seem to do aio_suspend()
as it should (it's never successfully blocked for me). I didn't have
much more luck with the signal structure, although that's conceivably
due to coding errors on my part. KAIO hasn't caused me one bit of
trouble (beyond the pending socket issue) and of course we know it
performs well (I haven't reached the stage where I've quantified that
yet).
--Chris
kaio-diffs.gz
Description: GNU Zip compressed data
|