]> bbs.cooldavid.org Git - net-next-2.6.git/commit - fs/xfs/xfs_dir2_block.c
[XFS] Radix tree based inode caching
authorDavid Chinner <dgc@sgi.com>
Tue, 28 Aug 2007 04:00:13 +0000 (14:00 +1000)
committerTim Shimmin <tes@chook.melbourne.sgi.com>
Mon, 15 Oct 2007 06:50:50 +0000 (16:50 +1000)
commitda353b0d64e070ae7c5342a0d56ec20ae9ef5cfb
tree84454023d649df67cc6b125c73746ddb341ac34e
parent39cd9f877e63ce7e02cdc7f5dbf1b908451c9532
[XFS] Radix tree based inode caching

One of the perpetual scaling problems XFS has is indexing it's incore
inodes. We currently uses hashes and the default hash sizes chosen can
only ever be a tradeoff between memory consumption and the maximum
realistic size of the cache.

As a result, anyone who has millions of inodes cached on a filesystem
needs to tunes the size of the cache via the ihashsize mount option to
allow decent scalability with inode cache operations.

A further problem is the separate inode cluster hash, whose size is based
on the ihashsize but is smaller, and so under certain conditions (sparse
cluster cache population) this can become a limitation long before the
inode hash is causing issues.

The following patchset removes the inode hash and cluster hash and
replaces them with radix trees to avoid the scalability limitations of the
hashes. It also reduces the size of the inodes by 3 pointers....

SGI-PV: 969561
SGI-Modid: xfs-linux-melb:xfs-kern:29481a

Signed-off-by: David Chinner <dgc@sgi.com>
Signed-off-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Tim Shimmin <tes@sgi.com>
20 files changed:
fs/xfs/linux-2.6/xfs_export.c
fs/xfs/xfs_ag.h
fs/xfs/xfs_buf_item.c
fs/xfs/xfs_clnt.h
fs/xfs/xfs_dir2_block.c
fs/xfs/xfs_dir2_data.c
fs/xfs/xfs_dir2_node.c
fs/xfs/xfs_dir2_sf.c
fs/xfs/xfs_error.c
fs/xfs/xfs_extfree_item.c
fs/xfs/xfs_iget.c
fs/xfs/xfs_inode.c
fs/xfs/xfs_inode.h
fs/xfs/xfs_mount.c
fs/xfs/xfs_mount.h
fs/xfs/xfs_rename.c
fs/xfs/xfs_trans_ail.c
fs/xfs/xfs_trans_extfree.c
fs/xfs/xfs_vfsops.c
fs/xfs/xfs_vnodeops.c