[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [Cluster-devel] [GFS2][Patch 0/10] GFS2 performance tweaks


On Tue, 2007-12-11 at 18:43 -0600, Bob Peterson wrote:
> Hi GFS2 Folks,
> The following is a set of ten patches designed to give gfs2 somewhat
> better performance than before.  Some are trivial and may or may not
> have a real impact on performance.  I'll let Steve Whitehouse decide
> which ones should go upstream and which ones aren't worth it.
> The patches may be summarized as follows:
I've applied most of these with the exceptions of:

> 1. Journal extent mapping
> 2. Get rid of useless "found" variable in quota.c
> 3. Run through full bitmaps quicker in gfs2_bitfit
> 4. Get rid of sd_statfs_mutex
 - Needs fixing as per previous email.

> 5. Shortcut in gfs2_write_alloc_required if writing past eof.
> 6. Reorganize function gfs2_glmutex_lock
> 7. Only fetch the dinode once in block_map
> 8. Only find indirect pointer buffers once in block_map
 - I'd really far rather see a fix to bz #307091. Also, although it
might make writes faster, I rather suspect that it will have the
opposite effect on reads, since the indirect blocks will not be in the
journal at all. So on balance I think I'd rather not take this patch
right now.

> 9. Move meta_inval to glops.c and declare static, more attach_bufdata
>    to trans.c and declare static.
 - This is also done in bz #349271 and I only want to apply one or the
other of these two patches so I'm deferring this until we know what we
are doing with the patches for #349271.

> 10. Function meta_read optimization.
> Regards,
> Bob Peterson
> Red Hat GFS

Also, please use git-send-email if you are sending multiple patches. I'm
not sure what you did to git to get the rather strange paths in these
patches, but I had to edit all the patches except the first one by hand
in order to get them to apply,


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]