[xfs-masters] [Bug 14208] XFS 5x slower than other filesystems when used as /var/tmp

[bugzilla-daemon at bugzilla.kernel.org]

http://bugzilla.kernel.org/show_bug.cgi?id=14208





--- Comment #5 from Eric Sandeen <sandeen at redhat.com>  2009-09-22 18:41:48 ---
"the same as reiserfs on microbenchmarks" is not necessarily the goal of xfs ;)

Anyway:

xfs_release()

                /*
                 * If we previously truncated this file and removed old data
                 * in the process, we want to initiate "early" writeout on
                 * the last close.  This is an attempt to combat the notorious
                 * NULL files problem which is particularly noticable from a
                 * truncate down, buffered (re-)write (delalloc), followed by
                 * a crash.  What we are effectively doing here is
                 * significantly reducing the time window where we'd otherwise
                 * be exposed to that problem.
                 */
                truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
                if (truncated && VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0)
                        xfs_flush_pages(ip, 0, -1, XFS_B_ASYNC, FI_NONE);

This essentially calls filemap_fdatawrite, which writes dirty data pages.

-- 
Configure bugmail: http://bugzilla.kernel.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.