diff options
| author | OGAWA Hirofumi <hirofumi@mail.parknet.co.jp> | 2013-12-04 20:34:29 +0900 |
|---|---|---|
| committer | Daniel Phillips <daniel@tux3.org> | 2013-12-04 20:34:29 +0900 |
| commit | 8e29513e2ceb618e6d9443138c7c7e522e32ec30 (patch) | |
| tree | f93681e5ceb377fb7541a118978d73fcf1df6485 | |
| parent | ec77ebbf001200342df46905c22aacc07680095e (diff) | |
| download | linux-tux3-8e29513e2ceb618e6d9443138c7c7e522e32ec30.tar.gz | |
tux3: Optimize split position for append write in dleaf2_write()
Append write is common operation, and if append write, split at center
is not optimal for leaf split.
So, this splits leaf at sentinel, if leaf is most right and key is >=
sentinel.
FIXME1: we should apply this for bnode too.
FIXME2: this changes split position only if most right leaf. Should we
split at sentinel by filling hole too?
Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
| -rw-r--r-- | fs/tux3/dleaf2.c | 15 |
1 files changed, 15 insertions, 0 deletions
diff --git a/fs/tux3/dleaf2.c b/fs/tux3/dleaf2.c index c1fdc04e2fe76..fa424c6635764 100644 --- a/fs/tux3/dleaf2.c +++ b/fs/tux3/dleaf2.c @@ -679,6 +679,21 @@ static int dleaf2_write(struct btree *btree, tuxkey_t key_bottom, return 0; need_split: + /* FIXME: do we should split at sentinel when filling hole? */ + if (key_limit == TUXKEY_LIMIT) { + struct diskextent2 *sentinel = + dleaf->table + be16_to_cpu(dleaf->count) - 1; + + /* If append write, split at sentinel */ + *split_hint = get_logical(sentinel); + if (key->start >= *split_hint) { + tux3_dbg("key %Lu bottom %Lu, limit %Lu, hint %Lu", + key->start, key_bottom, key_limit, + *split_hint); + return 1; + } + } + /* FIXME: use better split position */ *split_hint = dleaf2_split_at_center(dleaf); return 1; /* try to split dleaf2 */ |