Re: [PATCH v2] btrfs: use customized batch size for total_bytes_pinned
Omar Sandoval 於 2018-07-13 06:19 寫到:
On Wed, Jul 11, 2018 at 11:59:36PM +0800, Ethan Lien wrote:
In commit b150a4f10d878 ("Btrfs: use a percpu to keep track of
possibly
pinned bytes") we use total_bytes_pinned to track how many bytes we
are
going to free in this transaction. When we are close to ENOSPC, we
check it
and know if we can make the allocation by commit the current
transaction.
For every data/metadata extent we are going to free, we add
total_bytes_pinned in btrfs_free_extent() and btrfs_free_tree_block(),
and
release it in unpin_extent_range() when we finish the transaction. So
this
is a variable we frequently update but rarely read - just the suitable
use of percpu_counter. But in previous commit we update
total_bytes_pinned
by default 32 batch size, making every update essentially a spin lock
protected update. Since every spin lock/unlock operation involves
syncing
a globally used variable and some kind of barrier in a SMP system,
this is
more expensive than using total_bytes_pinned as a simple atomic64_t.
So
fix this by using a customized batch size. Since we only read
total_bytes_pinned when we are close to ENOSPC and fail to alloc new
chunk,
we can use a really large batch size and have nearly no penalty in
most
cases.
[Test]
We test the patch on a 4-cores x86 machine:
1. falloate a 16GiB size test file.
2. take snapshot (so all following writes will be cow write).
3. run a 180 sec, 4 jobs, 4K random write fio on test file.
We also add a temporary lockdep class on percpu_counter's spin lock
used
by total_bytes_pinned to track lock_stat.
[Results]
unpatched:
lock_stat version 0.4
---
class namecon-bouncescontentions
waittime-min waittime-max waittime-total waittime-avg
acq-bounces
acquisitions holdtime-min holdtime-max holdtime-total
holdtime-avg
total_bytes_pinned_percpu:82 82
0.21 0.61 29.46 0.36
298340
635973 0.09 11.01 173476.25
0.27
patched:
lock_stat version 0.4
---
class namecon-bouncescontentions
waittime-min waittime-max waittime-total waittime-avg
acq-bounces
acquisitions holdtime-min holdtime-max holdtime-total
holdtime-avg
total_bytes_pinned_percpu: 1 1
0.62 0.62 0.62 0.62
13601
31542 0.14 9.61 11016.90
0.35
[Analysis]
Since the spin lock only protect a single in-memory variable, the
contentions (number of lock acquisitions that had to wait) in both
unpatched and patched version are low. But when we see acquisitions
and
acq-bounces, we get much lower counts in patched version. Here the
most
important metric is acq-bounces. It means how many times the lock get
transferred between different cpus, so the patch can really recude
cacheline bouncing of spin lock (also the global counter of
percpu_counter)
in a SMP system.
Fixes: b150a4f10d878 ("Btrfs: use a percpu to keep track of possibly
pinned bytes")
Signed-off-by: Ethan Lien
---
V2:
Rewrite commit comments.
Add lock_stat test.
Pull dirty_metadata_bytes out to a separate patch.
fs/btrfs/ctree.h | 1 +
fs/btrfs/extent-tree.c | 46
--
2 files changed, 32 insertions(+), 15 deletions(-)
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 118346aceea9..df682a521635 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -422,6 +422,7 @@ struct btrfs_space_info {
* time the transaction commits.
*/
struct percpu_counter total_bytes_pinned;
+ s32 total_bytes_pinned_batch;
Can this just be a constant instead of adding it to space_info?
Yes constant is better here, I'll resend it, thanks.
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs"
in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH v2] btrfs: use customized batch size for total_bytes_pinned
On Wed, Jul 11, 2018 at 11:59:36PM +0800, Ethan Lien wrote:
> In commit b150a4f10d878 ("Btrfs: use a percpu to keep track of possibly
> pinned bytes") we use total_bytes_pinned to track how many bytes we are
> going to free in this transaction. When we are close to ENOSPC, we check it
> and know if we can make the allocation by commit the current transaction.
> For every data/metadata extent we are going to free, we add
> total_bytes_pinned in btrfs_free_extent() and btrfs_free_tree_block(), and
> release it in unpin_extent_range() when we finish the transaction. So this
> is a variable we frequently update but rarely read - just the suitable
> use of percpu_counter. But in previous commit we update total_bytes_pinned
> by default 32 batch size, making every update essentially a spin lock
> protected update. Since every spin lock/unlock operation involves syncing
> a globally used variable and some kind of barrier in a SMP system, this is
> more expensive than using total_bytes_pinned as a simple atomic64_t. So
> fix this by using a customized batch size. Since we only read
> total_bytes_pinned when we are close to ENOSPC and fail to alloc new chunk,
> we can use a really large batch size and have nearly no penalty in most
> cases.
>
>
> [Test]
> We test the patch on a 4-cores x86 machine:
> 1. falloate a 16GiB size test file.
> 2. take snapshot (so all following writes will be cow write).
> 3. run a 180 sec, 4 jobs, 4K random write fio on test file.
>
> We also add a temporary lockdep class on percpu_counter's spin lock used
> by total_bytes_pinned to track lock_stat.
>
>
> [Results]
> unpatched:
> lock_stat version 0.4
> ---
> class namecon-bouncescontentions
> waittime-min waittime-max waittime-total waittime-avgacq-bounces
> acquisitions holdtime-min holdtime-max holdtime-total holdtime-avg
>
>total_bytes_pinned_percpu:82 82
> 0.21 0.61 29.46 0.36 298340
> 635973 0.09 11.01 173476.25 0.27
>
>
> patched:
> lock_stat version 0.4
> ---
> class namecon-bouncescontentions
> waittime-min waittime-max waittime-total waittime-avgacq-bounces
> acquisitions holdtime-min holdtime-max holdtime-total holdtime-avg
>
>total_bytes_pinned_percpu: 1 1
> 0.62 0.62 0.62 0.62 13601
>31542 0.14 9.61 11016.90 0.35
>
>
> [Analysis]
> Since the spin lock only protect a single in-memory variable, the
> contentions (number of lock acquisitions that had to wait) in both
> unpatched and patched version are low. But when we see acquisitions and
> acq-bounces, we get much lower counts in patched version. Here the most
> important metric is acq-bounces. It means how many times the lock get
> transferred between different cpus, so the patch can really recude
> cacheline bouncing of spin lock (also the global counter of percpu_counter)
> in a SMP system.
>
> Fixes: b150a4f10d878 ("Btrfs: use a percpu to keep track of possibly
> pinned bytes")
>
> Signed-off-by: Ethan Lien
> ---
>
> V2:
> Rewrite commit comments.
> Add lock_stat test.
> Pull dirty_metadata_bytes out to a separate patch.
>
> fs/btrfs/ctree.h | 1 +
> fs/btrfs/extent-tree.c | 46 --
> 2 files changed, 32 insertions(+), 15 deletions(-)
>
> diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
> index 118346aceea9..df682a521635 100644
> --- a/fs/btrfs/ctree.h
> +++ b/fs/btrfs/ctree.h
> @@ -422,6 +422,7 @@ struct btrfs_space_info {
>* time the transaction commits.
>*/
> struct percpu_counter total_bytes_pinned;
> + s32 total_bytes_pinned_batch;
Can this just be a constant instead of adding it to space_info?
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH v2] btrfs: use customized batch size for total_bytes_pinned
Nikolay Borisov 於 2018-07-12 15:07 寫到:
On 11.07.2018 18:59, Ethan Lien wrote:
In commit b150a4f10d878 ("Btrfs: use a percpu to keep track of
possibly
pinned bytes") we use total_bytes_pinned to track how many bytes we
are
going to free in this transaction. When we are close to ENOSPC, we
check it
and know if we can make the allocation by commit the current
transaction.
For every data/metadata extent we are going to free, we add
total_bytes_pinned in btrfs_free_extent() and btrfs_free_tree_block(),
and
release it in unpin_extent_range() when we finish the transaction. So
this
is a variable we frequently update but rarely read - just the suitable
use of percpu_counter. But in previous commit we update
total_bytes_pinned
by default 32 batch size, making every update essentially a spin lock
protected update. Since every spin lock/unlock operation involves
syncing
a globally used variable and some kind of barrier in a SMP system,
this is
more expensive than using total_bytes_pinned as a simple atomic64_t.
So
fix this by using a customized batch size. Since we only read
total_bytes_pinned when we are close to ENOSPC and fail to alloc new
chunk,
we can use a really large batch size and have nearly no penalty in
most
cases.
[Test]
We test the patch on a 4-cores x86 machine:
1. falloate a 16GiB size test file.
2. take snapshot (so all following writes will be cow write).
3. run a 180 sec, 4 jobs, 4K random write fio on test file.
We also add a temporary lockdep class on percpu_counter's spin lock
used
by total_bytes_pinned to track lock_stat.
[Results]
unpatched:
lock_stat version 0.4
---
class namecon-bouncescontentions
waittime-min waittime-max waittime-total waittime-avg
acq-bounces
acquisitions holdtime-min holdtime-max holdtime-total
holdtime-avg
total_bytes_pinned_percpu:82 82
0.21 0.61 29.46 0.36
298340
635973 0.09 11.01 173476.25
0.27
patched:
lock_stat version 0.4
---
class namecon-bouncescontentions
waittime-min waittime-max waittime-total waittime-avg
acq-bounces
acquisitions holdtime-min holdtime-max holdtime-total
holdtime-avg
total_bytes_pinned_percpu: 1 1
0.62 0.62 0.62 0.62
13601
31542 0.14 9.61 11016.90
0.35
According to these numbers though, the best-case (waittime-min and the
average waittime-avg) have actually regressed. So what really saves you
In patched version we only got one contentions count so the min/max/avg
waittime all come from that count. I think the min/max/avg waittime here
are inconclusive since we have only one sample count so it could be
biased.
is the fact that the number of time we had to go to the best/average
case is reduced, due to the larger batch. I guess in that regard you
are
in the clear. Another pertinent question is did you observe any
significant impact on run times of actual workloads or, say,
transaction
commit times or anything like that? I think this case will only be hit
when the filesystem is struggling to satisfy a metadata allocation has
to cycle through all stages .
The frequently update of total_bytes_pinned can happen in normal cow
write path.
In our test, any re-write to a existing 4K data block will trigger cow
and we release
old data block by btrfs_free_extent() and in that time we add
total_bytes_pinned.
As the test file is written out, eventually almost every 4K write will
trigger a +4K and a
-4K update to total_bytes_pinned. But so far we haven't seen significant
performance
improvement from the test.
[Analysis]
Since the spin lock only protect a single in-memory variable, the
contentions (number of lock acquisitions that had to wait) in both
unpatched and patched version are low. But when we see acquisitions
and
acq-bounces, we get much lower counts in patched version. Here the
most
important metric is acq-bounces. It means how many times the lock get
transferred between different cpus, so the patch can really recude
nit: s/recude/reduce - no need to resend i guess david will fix it on
the way into the tree
cacheline bouncing of spin lock (also the global counter of
percpu_counter)
in a SMP system.
Fixes: b150a4f10d878 ("Btrfs: use a percpu to keep track of possibly
pinned bytes")
Signed-off-by: Ethan Lien
Reviewed-by: Nikolay Borisov
---
V2:
Rewrite commit comments.
Add lock_stat test.
Pull dirty_metadata_bytes out to a separate patch.
fs/btrfs/ctree.h | 1 +
fs/btrfs/extent-tree.c | 46
--
2 files change
Re: [PATCH v2] btrfs: use customized batch size for total_bytes_pinned
On 11.07.2018 18:59, Ethan Lien wrote:
> In commit b150a4f10d878 ("Btrfs: use a percpu to keep track of possibly
> pinned bytes") we use total_bytes_pinned to track how many bytes we are
> going to free in this transaction. When we are close to ENOSPC, we check it
> and know if we can make the allocation by commit the current transaction.
> For every data/metadata extent we are going to free, we add
> total_bytes_pinned in btrfs_free_extent() and btrfs_free_tree_block(), and
> release it in unpin_extent_range() when we finish the transaction. So this
> is a variable we frequently update but rarely read - just the suitable
> use of percpu_counter. But in previous commit we update total_bytes_pinned
> by default 32 batch size, making every update essentially a spin lock
> protected update. Since every spin lock/unlock operation involves syncing
> a globally used variable and some kind of barrier in a SMP system, this is
> more expensive than using total_bytes_pinned as a simple atomic64_t. So
> fix this by using a customized batch size. Since we only read
> total_bytes_pinned when we are close to ENOSPC and fail to alloc new chunk,
> we can use a really large batch size and have nearly no penalty in most
> cases.
>
>
> [Test]
> We test the patch on a 4-cores x86 machine:
> 1. falloate a 16GiB size test file.
> 2. take snapshot (so all following writes will be cow write).
> 3. run a 180 sec, 4 jobs, 4K random write fio on test file.
>
> We also add a temporary lockdep class on percpu_counter's spin lock used
> by total_bytes_pinned to track lock_stat.
>
>
> [Results]
> unpatched:
> lock_stat version 0.4
> ---
> class namecon-bouncescontentions
> waittime-min waittime-max waittime-total waittime-avgacq-bounces
> acquisitions holdtime-min holdtime-max holdtime-total holdtime-avg
>
>total_bytes_pinned_percpu:82 82
> 0.21 0.61 29.46 0.36 298340
> 635973 0.09 11.01 173476.25 0.27
>
>
> patched:
> lock_stat version 0.4
> ---
> class namecon-bouncescontentions
> waittime-min waittime-max waittime-total waittime-avgacq-bounces
> acquisitions holdtime-min holdtime-max holdtime-total holdtime-avg
>
>total_bytes_pinned_percpu: 1 1
> 0.62 0.62 0.62 0.62 13601
>31542 0.14 9.61 11016.90 0.35
>
According to these numbers though, the best-case (waittime-min and the
average waittime-avg) have actually regressed. So what really saves you
is the fact that the number of time we had to go to the best/average
case is reduced, due to the larger batch. I guess in that regard you are
in the clear. Another pertinent question is did you observe any
significant impact on run times of actual workloads or, say, transaction
commit times or anything like that? I think this case will only be hit
when the filesystem is struggling to satisfy a metadata allocation has
to cycle through all stages .
>
> [Analysis]
> Since the spin lock only protect a single in-memory variable, the
> contentions (number of lock acquisitions that had to wait) in both
> unpatched and patched version are low. But when we see acquisitions and
> acq-bounces, we get much lower counts in patched version. Here the most
> important metric is acq-bounces. It means how many times the lock get
> transferred between different cpus, so the patch can really recude
nit: s/recude/reduce - no need to resend i guess david will fix it on
the way into the tree
> cacheline bouncing of spin lock (also the global counter of percpu_counter)
> in a SMP system.
>
> Fixes: b150a4f10d878 ("Btrfs: use a percpu to keep track of possibly
> pinned bytes")
>
> Signed-off-by: Ethan Lien
Reviewed-by: Nikolay Borisov
> ---
>
> V2:
> Rewrite commit comments.
> Add lock_stat test.
> Pull dirty_metadata_bytes out to a separate patch.
>
> fs/btrfs/ctree.h | 1 +
> fs/btrfs/extent-tree.c | 46 --
> 2 files changed, 32 insertions(+), 15 deletions(-)
>
> diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
> index 118346aceea9..df682a521635 100644
> --- a/fs/btrfs/ctree.h
> +++ b/fs/btrfs/ctree.h
> @@ -422,6 +422,7 @@ struct btrfs_space_info {
>* time the transaction commits.
>*/
> struct percpu_counter total_bytes_pinned;
> + s32 total_bytes_pinned_batch;
>
> struct list_head list;
> /* Protected by the spinlock 'lock'. */
> diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
> index 3d9fe58c0080..937113534ef4 100644
> --- a/fs/btrfs/extent-tree.c
> +++ b/fs/btrfs/extent-tree.c
> @
