2006-10-26 11:59:45

by Holden Karau

[permalink] [raw]
Subject: [PATCH 1/1] fat: improve sync performance by grouping writes in fat_mirror_bhs [really unmangled]

From: Holden Karau <[email protected]> http://www.holdenkarau.com

This is an attempt at improving fat_mirror_bhs in sync mode [namely it
writes all of the data for a backup block, and then blocks untill
finished]. The old behaviour would write & block in smaller chunks, so
this should be slightly faster. It also removes the fixme requesting
that it be fixed to behave this way :-)
Signed-off-by: Holden Karau <[email protected]> http://www.holdenkarau.com
---
Sorry about the mangled patch, this time it really shouldn't be mangled :-)
Important note; I do not normally play with filesystems. This MAY eat
your file system, it has not eaten mine but that does not mean much at
all [although I don't think it will]. I'd greatly appreciate comments
& suggestions. In case the patch gets mangled [again] I've put it up at
http://www.holdenkarau.com/~holden/projects/fat/001_improve_fat_sync_performance.patch


And now for the actual patch:

--- a/fs/fat/fatent.c 2006-09-19 23:42:06.000000000 -0400
+++ b/fs/fat/fatent.c 2006-10-25 19:14:14.000000000 -0400
@@ -1,5 +1,6 @@
/*
* Copyright (C) 2004, OGAWA Hirofumi
+ * Copyright (C) 2006, Holden Karau [Xandros]
* Released under GPL v2.
*/

@@ -343,34 +344,46 @@ int fat_ent_read(struct inode *inode, st
return ops->ent_get(fatent);
}

-/* FIXME: We can write the blocks as more big chunk. */
static int fat_mirror_bhs(struct super_block *sb, struct buffer_head **bhs,
- int nr_bhs)
+ int nr_bhs ) {
+ return fat_mirror_bhs_optw(sb , bhs , nr_bhs, 0);
+}
+
+static int fat_mirror_bhs_optw(struct super_block *sb, struct buffer_head **bhs,
+ int nr_bhs , int wait)
{
struct msdos_sb_info *sbi = MSDOS_SB(sb);
- struct buffer_head *c_bh;
+ struct buffer_head *c_bh[nr_bhs];
int err, n, copy;

+ /* Always wait if mounted -o sync */
+ if (sb->s_flags & MS_SYNCHRONOUS ) {
+ wait = 1;
+ }
+
err = 0;
+ err = fat_sync_bhs_optw( bhs , nr_bhs , wait);
+ if (err)
+ goto error;
for (copy = 1; copy < sbi->fats; copy++) {
sector_t backup_fat = sbi->fat_length * copy;
-
for (n = 0; n < nr_bhs; n++) {
- c_bh = sb_getblk(sb, backup_fat + bhs[n]->b_blocknr);
- if (!c_bh) {
+ c_bh[n] = sb_getblk(sb, backup_fat + bhs[n]->b_blocknr);
+ if (!c_bh[n]) {
err = -ENOMEM;
goto error;
}
- memcpy(c_bh->b_data, bhs[n]->b_data, sb->s_blocksize);
- set_buffer_uptodate(c_bh);
- mark_buffer_dirty(c_bh);
- if (sb->s_flags & MS_SYNCHRONOUS)
- err = sync_dirty_buffer(c_bh);
- brelse(c_bh);
+ set_buffer_uptodate(c_bh[n]);
+ mark_buffer_dirty(c_bh[n]);
+ memcpy(c_bh[n]->b_data, bhs[n]->b_data, sb->s_blocksize);
+ }
+ err = fat_sync_bhs_optw( c_bh , nr_bhs , wait );
+ for (n = 0; n < nr_bhs; n++ ) {
+ brelse(c_bh[n]);
+ }
if (err)
goto error;
}
- }
error:
return err;
}
@@ -383,12 +396,7 @@ int fat_ent_write(struct inode *inode, s
int err;

ops->ent_put(fatent, new);
- if (wait) {
- err = fat_sync_bhs(fatent->bhs, fatent->nr_bhs);
- if (err)
- return err;
- }
- return fat_mirror_bhs(sb, fatent->bhs, fatent->nr_bhs);
+ return fat_mirror_bhs_optw(sb, fatent->bhs, fatent->nr_bhs , wait);
}

static inline int fat_ent_next(struct msdos_sb_info *sbi,
@@ -505,9 +513,9 @@ out:
fatent_brelse(&fatent);
if (!err) {
if (inode_needs_sync(inode))
- err = fat_sync_bhs(bhs, nr_bhs);
- if (!err)
- err = fat_mirror_bhs(sb, bhs, nr_bhs);
+ err = fat_mirror_bhs_optw(sb , bhs, nr_bhs , 1);
+ else
+ err = fat_mirror_bhs_optw(sb, bhs, nr_bhs , 0 );
}
for (i = 0; i < nr_bhs; i++)
brelse(bhs[i]);
@@ -549,11 +557,6 @@ int fat_free_clusters(struct inode *inod
}

if (nr_bhs + fatent.nr_bhs > MAX_BUF_PER_PAGE) {
- if (sb->s_flags & MS_SYNCHRONOUS) {
- err = fat_sync_bhs(bhs, nr_bhs);
- if (err)
- goto error;
- }
err = fat_mirror_bhs(sb, bhs, nr_bhs);
if (err)
goto error;
@@ -564,11 +567,6 @@ int fat_free_clusters(struct inode *inod
fat_collect_bhs(bhs, &nr_bhs, &fatent);
} while (cluster != FAT_ENT_EOF);

- if (sb->s_flags & MS_SYNCHRONOUS) {
- err = fat_sync_bhs(bhs, nr_bhs);
- if (err)
- goto error;
- }
err = fat_mirror_bhs(sb, bhs, nr_bhs);
error:
fatent_brelse(&fatent);
--- a/fs/fat/misc.c 2006-09-19 23:42:06.000000000 -0400
+++ b/fs/fat/misc.c 2006-10-25 18:54:27.000000000 -0400
@@ -194,11 +194,17 @@ void fat_date_unix2dos(int unix_date, __

EXPORT_SYMBOL_GPL(fat_date_unix2dos);

-int fat_sync_bhs(struct buffer_head **bhs, int nr_bhs)
+
+int fat_sync_bhs(struct buffer_head **bhs, int nr_bhs ) {
+ return fat_sync_bhs_optw(bhs , nr_bhs , 1);
+}
+
+int fat_sync_bhs_optw(struct buffer_head **bhs, int nr_bhs ,int wait)
{
int i, err = 0;

ll_rw_block(SWRITE, nr_bhs, bhs);
+ if (wait) {
for (i = 0; i < nr_bhs; i++) {
wait_on_buffer(bhs[i]);
if (buffer_eopnotsupp(bhs[i])) {
@@ -207,6 +213,7 @@ int fat_sync_bhs(struct buffer_head **bh
} else if (!err && !buffer_uptodate(bhs[i]))
err = -EIO;
}
+ }
return err;
}

--- a/include/linux/msdos_fs.h 2006-09-19 23:42:06.000000000 -0400
+++ b/include/linux/msdos_fs.h 2006-10-25 18:53:50.000000000 -0400
@@ -419,6 +419,7 @@ extern int fat_chain_add(struct inode *i
extern int date_dos2unix(unsigned short time, unsigned short date);
extern void fat_date_unix2dos(int unix_date, __le16 *time, __le16 *date);
extern int fat_sync_bhs(struct buffer_head **bhs, int nr_bhs);
+extern int fat_sync_bhs_optw(struct buffer_head **bhs, int nr_bhs, int wait);

int fat_cache_init(void);
void fat_cache_destroy(void);


2006-10-26 12:29:26

by Nick Piggin

[permalink] [raw]
Subject: Re: [PATCH 1/1] fat: improve sync performance by grouping writes in fat_mirror_bhs [really unmangled]

Holden Karau wrote:
> From: Holden Karau <[email protected]> http://www.holdenkarau.com
>
> This is an attempt at improving fat_mirror_bhs in sync mode [namely it
> writes all of the data for a backup block, and then blocks untill
> finished]. The old behaviour would write & block in smaller chunks, so
> this should be slightly faster. It also removes the fixme requesting
> that it be fixed to behave this way :-)
> Signed-off-by: Holden Karau <[email protected]> http://www.holdenkarau.com

So how much is performance improved?

--
SUSE Labs, Novell Inc.
Send instant messages to your online friends http://au.messenger.yahoo.com

2006-10-26 15:31:38

by Jörn Engel

[permalink] [raw]
Subject: Re: [PATCH 1/1] fat: improve sync performance by grouping writes in fat_mirror_bhs [really unmangled]

I didn't pay too much attention, but found some low hanging fruits.

On Thu, 26 October 2006 07:59:42 -0400, Holden Karau wrote:
>
> -/* FIXME: We can write the blocks as more big chunk. */
> static int fat_mirror_bhs(struct super_block *sb, struct buffer_head **bhs,
> - int nr_bhs)
> + int nr_bhs ) {
> + return fat_mirror_bhs_optw(sb , bhs , nr_bhs, 0);
> +}
> +
> +static int fat_mirror_bhs_optw(struct super_block *sb, struct buffer_head **bhs,
> + int nr_bhs , int wait)

Does this compile without warnings? Looks as if you should reverse
the order of the two functions.

> {
> struct msdos_sb_info *sbi = MSDOS_SB(sb);
> - struct buffer_head *c_bh;
> + struct buffer_head *c_bh[nr_bhs];
> int err, n, copy;
>
> + /* Always wait if mounted -o sync */
> + if (sb->s_flags & MS_SYNCHRONOUS ) {
> + wait = 1;
> + }

Coding style. Use a tab for indentation and don't use braces for
single-line conditional statements.

> +
> err = 0;
> + err = fat_sync_bhs_optw( bhs , nr_bhs , wait);

The err=0; is superfluous now, isn't it?

> + if (err)
> + goto error;

Indentation.

J?rn

--
Fantasy is more important than knowledge. Knowledge is limited,
while fantasy embraces the whole world.
-- Albert Einstein

2006-10-27 18:55:11

by Holden Karau

[permalink] [raw]
Subject: Re: [PATCH 1/1] fat: improve sync performance by grouping writes in fat_mirror_bhs [really unmangled]

Based on some rather off-the cuff bench marking I've done I get around
a 5 to 10% performance improve over the stock 2.6.18.1 performance. As
the saying goes, there are lies, damn lies, and bench marks, so YMMV.

On 10/26/06, Nick Piggin <[email protected]> wrote:
> Holden Karau wrote:
> > From: Holden Karau <[email protected]> http://www.holdenkarau.com
> >
> > This is an attempt at improving fat_mirror_bhs in sync mode [namely it
> > writes all of the data for a backup block, and then blocks untill
> > finished]. The old behaviour would write & block in smaller chunks, so
> > this should be slightly faster. It also removes the fixme requesting
> > that it be fixed to behave this way :-)
> > Signed-off-by: Holden Karau <[email protected]> http://www.holdenkarau.com
>
> So how much is performance improved?
>
> --
> SUSE Labs, Novell Inc.
> Send instant messages to your online friends http://au.messenger.yahoo.com
>
>


--
Cell: 613-276-1645

2006-10-27 18:56:21

by Holden Karau

[permalink] [raw]
Subject: Re: [PATCH 1/1] fat: improve sync performance by grouping writes in fat_mirror_bhs [really unmangled]

Hi J?rn,

Thanks for your time, I'll make those changes [along with a few other
things I noticed while benchmarking it]. Before I put together a
patch, does anyone else see any obvious stuff I should clean up?

Cheers,

Holden :-)

On 10/26/06, J?rn Engel <[email protected]> wrote:
> I didn't pay too much attention, but found some low hanging fruits.
>
> On Thu, 26 October 2006 07:59:42 -0400, Holden Karau wrote:
> >
> > -/* FIXME: We can write the blocks as more big chunk. */
> > static int fat_mirror_bhs(struct super_block *sb, struct buffer_head **bhs,
> > - int nr_bhs)
> > + int nr_bhs ) {
> > + return fat_mirror_bhs_optw(sb , bhs , nr_bhs, 0);
> > +}
> > +
> > +static int fat_mirror_bhs_optw(struct super_block *sb, struct buffer_head **bhs,
> > + int nr_bhs , int wait)
>
> Does this compile without warnings? Looks as if you should reverse
> the order of the two functions.
>
For some reason it compiles without warnings for me, but I'll switch the order.
> > {
> > struct msdos_sb_info *sbi = MSDOS_SB(sb);
> > - struct buffer_head *c_bh;
> > + struct buffer_head *c_bh[nr_bhs];
> > int err, n, copy;
> >
> > + /* Always wait if mounted -o sync */
> > + if (sb->s_flags & MS_SYNCHRONOUS ) {
> > + wait = 1;
> > + }
>
> Coding style. Use a tab for indentation and don't use braces for
> single-line conditional statements.
>
Sorry about that. A lot of the places where I used braces are because
I had some debugging output in there while I was hacking on it. I'll
change it.
> > +
> > err = 0;
> > + err = fat_sync_bhs_optw( bhs , nr_bhs , wait);
>
> The err=0; is superfluous now, isn't it?
>
.... no comment :-)
> > + if (err)
> > + goto error;
>
> Indentation.
>
oops :-) I'll fix that.
> J?rn
>
> --
> Fantasy is more important than knowledge. Knowledge is limited,
> while fantasy embraces the whole world.
> -- Albert Einstein
>


--
Cell: 613-276-1645

2006-10-27 21:01:32

by Josef Sipek

[permalink] [raw]
Subject: Re: [PATCH 1/1] fat: improve sync performance by grouping writes in fat_mirror_bhs [really unmangled]

Some of these may overlap with what others suggested.

On Thu, Oct 26, 2006 at 07:59:42AM -0400, Holden Karau wrote:
...
> --- a/fs/fat/fatent.c 2006-09-19 23:42:06.000000000 -0400
> +++ b/fs/fat/fatent.c 2006-10-25 19:14:14.000000000 -0400
...
> -/* FIXME: We can write the blocks as more big chunk. */
> static int fat_mirror_bhs(struct super_block *sb, struct buffer_head **bhs,
> - int nr_bhs)
> + int nr_bhs ) {
> + return fat_mirror_bhs_optw(sb , bhs , nr_bhs, 0);
> +}

Coding style... (brackets and indentation)

> +static int fat_mirror_bhs_optw(struct super_block *sb, struct buffer_head **bhs,
> + int nr_bhs , int wait)
> {
> struct msdos_sb_info *sbi = MSDOS_SB(sb);
> - struct buffer_head *c_bh;
> + struct buffer_head *c_bh[nr_bhs];
> int err, n, copy;
>
> + /* Always wait if mounted -o sync */
> + if (sb->s_flags & MS_SYNCHRONOUS ) {
> + wait = 1;
> + }

Ditto (indentation).

> err = 0;
> + err = fat_sync_bhs_optw( bhs , nr_bhs , wait);
> + if (err)
> + goto error;
> for (copy = 1; copy < sbi->fats; copy++) {
> sector_t backup_fat = sbi->fat_length * copy;
> -
> for (n = 0; n < nr_bhs; n++) {
> - c_bh = sb_getblk(sb, backup_fat + bhs[n]->b_blocknr);
> - if (!c_bh) {
> + c_bh[n] = sb_getblk(sb, backup_fat + bhs[n]->b_blocknr);
> + if (!c_bh[n]) {
> err = -ENOMEM;
> goto error;
> }
> - memcpy(c_bh->b_data, bhs[n]->b_data, sb->s_blocksize);
> - set_buffer_uptodate(c_bh);
> - mark_buffer_dirty(c_bh);
> - if (sb->s_flags & MS_SYNCHRONOUS)
> - err = sync_dirty_buffer(c_bh);
> - brelse(c_bh);
> + set_buffer_uptodate(c_bh[n]);
> + mark_buffer_dirty(c_bh[n]);
> + memcpy(c_bh[n]->b_data, bhs[n]->b_data, sb->s_blocksize);
> + }
> + err = fat_sync_bhs_optw( c_bh , nr_bhs , wait );
> + for (n = 0; n < nr_bhs; n++ ) {
> + brelse(c_bh[n]);
> + }
> if (err)
> goto error;
> }
> - }
> error:
> return err;
> }

Ditto.

> @@ -505,9 +513,9 @@ out:
> fatent_brelse(&fatent);
> if (!err) {
> if (inode_needs_sync(inode))
> - err = fat_sync_bhs(bhs, nr_bhs);
> - if (!err)
> - err = fat_mirror_bhs(sb, bhs, nr_bhs);
> + err = fat_mirror_bhs_optw(sb , bhs, nr_bhs , 1);
> + else
> + err = fat_mirror_bhs_optw(sb, bhs, nr_bhs , 0 );

Ditto.

> --- a/fs/fat/misc.c 2006-09-19 23:42:06.000000000 -0400
> +++ b/fs/fat/misc.c 2006-10-25 18:54:27.000000000 -0400
> @@ -194,11 +194,17 @@ void fat_date_unix2dos(int unix_date, __
>
> EXPORT_SYMBOL_GPL(fat_date_unix2dos);
>
> -int fat_sync_bhs(struct buffer_head **bhs, int nr_bhs)
> +
> +int fat_sync_bhs(struct buffer_head **bhs, int nr_bhs ) {
> + return fat_sync_bhs_optw(bhs , nr_bhs , 1);
> +}

Indentation & bracket placement.

> +int fat_sync_bhs_optw(struct buffer_head **bhs, int nr_bhs ,int wait)
> {
> int i, err = 0;
>
> ll_rw_block(SWRITE, nr_bhs, bhs);
> + if (wait) {
> for (i = 0; i < nr_bhs; i++) {
> wait_on_buffer(bhs[i]);
> if (buffer_eopnotsupp(bhs[i])) {
> @@ -207,6 +213,7 @@ int fat_sync_bhs(struct buffer_head **bh
> } else if (!err && !buffer_uptodate(bhs[i]))
> err = -EIO;
> }
> + }

Indentation.

Josef "Jeff" Sipek.

--
Mankind invented the atomic bomb, but no mouse would ever construct a
mousetrap.
- Albert Einstein