lkml.org 
[lkml]   [2002]   [Mar]   [1]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
Subject[patch] multipage pagecache writeout

These patches:

http://www.zip.com.au/~akpm/linux/patches/2.5/2.5.6-pre1/mpio-10-biobits.patch
http://www.zip.com.au/~akpm/linux/patches/2.5/2.5.6-pre1/mpio-20-core.patch
http://www.zip.com.au/~akpm/linux/patches/2.5/2.5.6-pre1/mpio-30-ext2.patch

implement multipage writeout from the pagecache. These patches require
the allocate-on-flush patches. The dalloc-30-ratcache patch is not a
requirement for the mpio series. But is recommended for
balls-to-the-wall how-fast-can-it-go testing.

Pages from the pagecache are given a disk mapping, are assembled into
large BIOs (up to half a megabyte) and these BIOs are injected direct
into the request layer.

These pages never have attached buffer_heads. The buffer layer is
completely bypassed for all write(2) data. As is, to some extent, the
request merging layer.

This patch should bypass the lru_list_lock contention problem, and the
ZONE_NORMAL-full-of-buffer_heads bug. (Well, this may require
multipage reads, too).



Future work includes:

- Implement buffer_head-less block_truncate_page().

- multipage reads.

A bit of a no-brainer, but first the current readahead code needs a
big shakeout.


Two additional patches are available:

http://www.zip.com.au/~akpm/linux/patches/2.5/2.5.6-pre1/tuning-10-request.patch

- The get_request starvation fix for 2.5.

This patch also increases the request queue by a
lot. Which implies that we can have as much as 512 megabytes
of I/O underway per device. This may sound excessive, but
the locked- and dirty-page accounting in the delalloc patch
only permits this to happen if the machine is large enough to
cope with it.

http://www.zip.com.au/~akpm/linux/patches/2.5/2.5.6-pre1/tuning-20-ext2-preread-inode.patch
- Pull the backing block for a new ext2 inode into
the buffercache when the inode is created. This fixes a
significant throughput problem with many-file writeout, where
the writer is continually interrupted by having to perform
reads.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:21    [W:0.023 / U:0.364 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site