The coming patchset is a new take on the old issue: ZRAM can currently be used only with zsmalloc even though this may not be the optimal combination for some configurations. The previous (unsuccessful) attempt dates back to 2015 [1] and is notable for the heated discussions it has caused.
The patchset in [1] had basically the only goal of enabling ZRAM/zbud combo which had a very narrow use case. Things have changed substantially since then, and now, with z3fold used widely as a zswap backend, I, as the z3fold maintainer, am getting requests to re-interate on making it possible to use ZRAM with any zpool-compatible backend, first of all z3fold.
The preliminary results for this work have been delivered at Linux Plumbers this year [2]. The talk at LPC, though having attracted limited interest, ended in a consensus to continue the work and pursue the goal of decoupling ZRAM from zsmalloc.
The current patchset has been stress tested on arm64 and x86_64 devices, including the Dell laptop I'm writing this message on now, not to mention several QEmu confugirations.
[1] https://lkml.org/lkml/2015/9/14/356
[2] https://linuxplumbersconf.org/event/4/contributions/551/
This patch adds the following functions to the zpool API:
- zpool_compact()
- zpool_get_num_compacted()
- zpool_huge_class_size()
The first one triggers compaction for the underlying allocator, the
second retrieves the number of pages migrated due to compaction for
the whole time of this pool's existence and the third one returns
the huge class size.
This API extension is done to align zpool API with zsmalloc API.
Signed-off-by: Vitaly Wool <[email protected]>
---
include/linux/zpool.h | 14 +++++++++++++-
mm/zpool.c | 36 ++++++++++++++++++++++++++++++++++++
2 files changed, 49 insertions(+), 1 deletion(-)
diff --git a/include/linux/zpool.h b/include/linux/zpool.h
index 51bf43076165..31f0c1360569 100644
--- a/include/linux/zpool.h
+++ b/include/linux/zpool.h
@@ -61,8 +61,13 @@ void *zpool_map_handle(struct zpool *pool, unsigned long handle,
void zpool_unmap_handle(struct zpool *pool, unsigned long handle);
+unsigned long zpool_compact(struct zpool *pool);
+
+unsigned long zpool_get_num_compacted(struct zpool *pool);
+
u64 zpool_get_total_size(struct zpool *pool);
+size_t zpool_huge_class_size(struct zpool *zpool);
/**
* struct zpool_driver - driver implementation for zpool
@@ -75,7 +80,10 @@ u64 zpool_get_total_size(struct zpool *pool);
* @shrink: shrink the pool.
* @map: map a handle.
* @unmap: unmap a handle.
- * @total_size: get total size of a pool.
+ * @compact: try to run compaction over a pool
+ * @get_num_compacted: get amount of compacted pages for a pool
+ * @total_size: get total size of a pool
+ * @huge_class_size: huge class threshold for pool pages.
*
* This is created by a zpool implementation and registered
* with zpool.
@@ -104,7 +112,11 @@ struct zpool_driver {
enum zpool_mapmode mm);
void (*unmap)(void *pool, unsigned long handle);
+ unsigned long (*compact)(void *pool);
+ unsigned long (*get_num_compacted)(void *pool);
+
u64 (*total_size)(void *pool);
+ size_t (*huge_class_size)(void *pool);
};
void zpool_register_driver(struct zpool_driver *driver);
diff --git a/mm/zpool.c b/mm/zpool.c
index 863669212070..55e69213c2eb 100644
--- a/mm/zpool.c
+++ b/mm/zpool.c
@@ -362,6 +362,30 @@ void zpool_unmap_handle(struct zpool *zpool, unsigned long handle)
zpool->driver->unmap(zpool->pool, handle);
}
+ /**
+ * zpool_compact() - try to run compaction over zpool
+ * @pool The zpool to compact
+ *
+ * Returns: the number of migrated pages
+ */
+unsigned long zpool_compact(struct zpool *zpool)
+{
+ return zpool->driver->compact ? zpool->driver->compact(zpool->pool) : 0;
+}
+
+
+/**
+ * zpool_get_num_compacted() - get the number of migrated/compacted pages
+ * @pool The zpool to get compaction statistic for
+ *
+ * Returns: the total number of migrated pages for the pool
+ */
+unsigned long zpool_get_num_compacted(struct zpool *zpool)
+{
+ return zpool->driver->get_num_compacted ?
+ zpool->driver->get_num_compacted(zpool->pool) : 0;
+}
+
/**
* zpool_get_total_size() - The total size of the pool
* @zpool: The zpool to check
@@ -375,6 +399,18 @@ u64 zpool_get_total_size(struct zpool *zpool)
return zpool->driver->total_size(zpool->pool);
}
+/**
+ * zpool_huge_class_size() - get size for the "huge" class
+ * @pool The zpool to check
+ *
+ * Returns: size of the huge class
+ */
+size_t zpool_huge_class_size(struct zpool *zpool)
+{
+ return zpool->driver->huge_class_size ?
+ zpool->driver->huge_class_size(zpool->pool) : 0;
+}
+
/**
* zpool_evictable() - Test if zpool is potentially evictable
* @zpool: The zpool to test
--
2.20.1
Add compaction callbacks for zpool compaction API extension.
Add huge_class_size callback too to be fully aligned.
With these in place, we can proceed with ZRAM modification
to use the universal (zpool) API.
Signed-off-by: Vitaly Wool <[email protected]>
---
mm/zsmalloc.c | 21 +++++++++++++++++++++
1 file changed, 21 insertions(+)
diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c
index 2b2b9aae8a3c..43f43272b998 100644
--- a/mm/zsmalloc.c
+++ b/mm/zsmalloc.c
@@ -437,11 +437,29 @@ static void zs_zpool_unmap(void *pool, unsigned long handle)
zs_unmap_object(pool, handle);
}
+static unsigned long zs_zpool_compact(void *pool)
+{
+ return zs_compact(pool);
+}
+
+static unsigned long zs_zpool_get_compacted(void *pool)
+{
+ struct zs_pool_stats stats;
+
+ zs_pool_stats(pool, &stats);
+ return stats.pages_compacted;
+}
+
static u64 zs_zpool_total_size(void *pool)
{
return zs_get_total_pages(pool) << PAGE_SHIFT;
}
+static size_t zs_zpool_huge_class_size(void *pool)
+{
+ return zs_huge_class_size(pool);
+}
+
static struct zpool_driver zs_zpool_driver = {
.type = "zsmalloc",
.owner = THIS_MODULE,
@@ -453,6 +471,9 @@ static struct zpool_driver zs_zpool_driver = {
.map = zs_zpool_map,
.unmap = zs_zpool_unmap,
.total_size = zs_zpool_total_size,
+ .compact = zs_zpool_compact,
+ .get_num_compacted = zs_zpool_get_compacted,
+ .huge_class_size = zs_zpool_huge_class_size,
};
MODULE_ALIAS("zpool-zsmalloc");
--
2.20.1
Change ZRAM into using zpool API. This patch allows to use any
zpool compatible allocation backend with ZRAM. It is meant to make
no functional changes to ZRAM.
zpool-registered backend can be selected via the module parameter
or kernel boot string. 'zsmalloc' is taken by default.
Signed-off-by: Vitaly Wool <[email protected]>
---
drivers/block/zram/Kconfig | 3 ++-
drivers/block/zram/zram_drv.c | 64 +++++++++++++++++++----------------
drivers/block/zram/zram_drv.h | 4 +--
3 files changed, 39 insertions(+), 32 deletions(-)
diff --git a/drivers/block/zram/Kconfig b/drivers/block/zram/Kconfig
index fe7a4b7d30cf..7248d5aa3468 100644
--- a/drivers/block/zram/Kconfig
+++ b/drivers/block/zram/Kconfig
@@ -1,8 +1,9 @@
# SPDX-License-Identifier: GPL-2.0
config ZRAM
tristate "Compressed RAM block device support"
- depends on BLOCK && SYSFS && ZSMALLOC && CRYPTO
+ depends on BLOCK && SYSFS && CRYPTO
select CRYPTO_LZO
+ select ZPOOL
help
Creates virtual block devices called /dev/zramX (X = 0, 1, ...).
Pages written to these disks are compressed and stored in memory
diff --git a/drivers/block/zram/zram_drv.c b/drivers/block/zram/zram_drv.c
index d58a359a6622..881f10f99a5d 100644
--- a/drivers/block/zram/zram_drv.c
+++ b/drivers/block/zram/zram_drv.c
@@ -43,6 +43,9 @@ static DEFINE_MUTEX(zram_index_mutex);
static int zram_major;
static const char *default_compressor = "lzo-rle";
+#define BACKEND_PAR_BUF_SIZE 32
+static char backend_par_buf[BACKEND_PAR_BUF_SIZE];
+
/* Module params (documentation at end) */
static unsigned int num_devices = 1;
/*
@@ -277,7 +280,7 @@ static ssize_t mem_used_max_store(struct device *dev,
down_read(&zram->init_lock);
if (init_done(zram)) {
atomic_long_set(&zram->stats.max_used_pages,
- zs_get_total_pages(zram->mem_pool));
+ zpool_get_total_size(zram->mem_pool) >> PAGE_SHIFT);
}
up_read(&zram->init_lock);
@@ -1020,7 +1023,7 @@ static ssize_t compact_store(struct device *dev,
return -EINVAL;
}
- zs_compact(zram->mem_pool);
+ zpool_compact(zram->mem_pool);
up_read(&zram->init_lock);
return len;
@@ -1048,17 +1051,14 @@ static ssize_t mm_stat_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
struct zram *zram = dev_to_zram(dev);
- struct zs_pool_stats pool_stats;
u64 orig_size, mem_used = 0;
- long max_used;
+ long max_used, num_compacted = 0;
ssize_t ret;
- memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
-
down_read(&zram->init_lock);
if (init_done(zram)) {
- mem_used = zs_get_total_pages(zram->mem_pool);
- zs_pool_stats(zram->mem_pool, &pool_stats);
+ mem_used = zpool_get_total_size(zram->mem_pool);
+ num_compacted = zpool_get_num_compacted(zram->mem_pool);
}
orig_size = atomic64_read(&zram->stats.pages_stored);
@@ -1068,11 +1068,11 @@ static ssize_t mm_stat_show(struct device *dev,
"%8llu %8llu %8llu %8lu %8ld %8llu %8lu %8llu\n",
orig_size << PAGE_SHIFT,
(u64)atomic64_read(&zram->stats.compr_data_size),
- mem_used << PAGE_SHIFT,
+ mem_used,
zram->limit_pages << PAGE_SHIFT,
max_used << PAGE_SHIFT,
(u64)atomic64_read(&zram->stats.same_pages),
- pool_stats.pages_compacted,
+ num_compacted,
(u64)atomic64_read(&zram->stats.huge_pages));
up_read(&zram->init_lock);
@@ -1133,27 +1133,30 @@ static void zram_meta_free(struct zram *zram, u64 disksize)
for (index = 0; index < num_pages; index++)
zram_free_page(zram, index);
- zs_destroy_pool(zram->mem_pool);
+ zpool_destroy_pool(zram->mem_pool);
vfree(zram->table);
}
static bool zram_meta_alloc(struct zram *zram, u64 disksize)
{
size_t num_pages;
+ char *backend;
num_pages = disksize >> PAGE_SHIFT;
zram->table = vzalloc(array_size(num_pages, sizeof(*zram->table)));
if (!zram->table)
return false;
- zram->mem_pool = zs_create_pool(zram->disk->disk_name);
+ backend = strlen(backend_par_buf) ? backend_par_buf : "zsmalloc";
+ zram->mem_pool = zpool_create_pool(backend, zram->disk->disk_name,
+ GFP_NOIO, NULL);
if (!zram->mem_pool) {
vfree(zram->table);
return false;
}
if (!huge_class_size)
- huge_class_size = zs_huge_class_size(zram->mem_pool);
+ huge_class_size = zpool_huge_class_size(zram->mem_pool);
return true;
}
@@ -1197,7 +1200,7 @@ static void zram_free_page(struct zram *zram, size_t index)
if (!handle)
return;
- zs_free(zram->mem_pool, handle);
+ zpool_free(zram->mem_pool, handle);
atomic64_sub(zram_get_obj_size(zram, index),
&zram->stats.compr_data_size);
@@ -1246,7 +1249,7 @@ static int __zram_bvec_read(struct zram *zram, struct page *page, u32 index,
size = zram_get_obj_size(zram, index);
- src = zs_map_object(zram->mem_pool, handle, ZS_MM_RO);
+ src = zpool_map_handle(zram->mem_pool, handle, ZPOOL_MM_RO);
if (size == PAGE_SIZE) {
dst = kmap_atomic(page);
memcpy(dst, src, PAGE_SIZE);
@@ -1260,7 +1263,7 @@ static int __zram_bvec_read(struct zram *zram, struct page *page, u32 index,
kunmap_atomic(dst);
zcomp_stream_put(zram->comp);
}
- zs_unmap_object(zram->mem_pool, handle);
+ zpool_unmap_handle(zram->mem_pool, handle);
zram_slot_unlock(zram, index);
/* Should NEVER happen. Return bio error if it does. */
@@ -1335,7 +1338,7 @@ static int __zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
if (unlikely(ret)) {
zcomp_stream_put(zram->comp);
pr_err("Compression failed! err=%d\n", ret);
- zs_free(zram->mem_pool, handle);
+ zpool_free(zram->mem_pool, handle);
return ret;
}
@@ -1354,33 +1357,34 @@ static int __zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
* if we have a 'non-null' handle here then we are coming
* from the slow path and handle has already been allocated.
*/
- if (!handle)
- handle = zs_malloc(zram->mem_pool, comp_len,
+ if (handle == 0)
+ ret = zpool_malloc(zram->mem_pool, comp_len,
__GFP_KSWAPD_RECLAIM |
__GFP_NOWARN |
__GFP_HIGHMEM |
- __GFP_MOVABLE);
- if (!handle) {
+ __GFP_MOVABLE,
+ &handle);
+ if (ret) {
zcomp_stream_put(zram->comp);
atomic64_inc(&zram->stats.writestall);
- handle = zs_malloc(zram->mem_pool, comp_len,
- GFP_NOIO | __GFP_HIGHMEM |
- __GFP_MOVABLE);
- if (handle)
+ ret = zpool_malloc(zram->mem_pool, comp_len,
+ GFP_NOIO | __GFP_HIGHMEM | __GFP_MOVABLE,
+ &handle);
+ if (ret == 0)
goto compress_again;
return -ENOMEM;
}
- alloced_pages = zs_get_total_pages(zram->mem_pool);
+ alloced_pages = zpool_get_total_size(zram->mem_pool) >> PAGE_SHIFT;
update_used_max(zram, alloced_pages);
if (zram->limit_pages && alloced_pages > zram->limit_pages) {
zcomp_stream_put(zram->comp);
- zs_free(zram->mem_pool, handle);
+ zpool_free(zram->mem_pool, handle);
return -ENOMEM;
}
- dst = zs_map_object(zram->mem_pool, handle, ZS_MM_WO);
+ dst = zpool_map_handle(zram->mem_pool, handle, ZPOOL_MM_WO);
src = zstrm->buffer;
if (comp_len == PAGE_SIZE)
@@ -1390,7 +1394,7 @@ static int __zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
kunmap_atomic(src);
zcomp_stream_put(zram->comp);
- zs_unmap_object(zram->mem_pool, handle);
+ zpool_unmap_handle(zram->mem_pool, handle);
atomic64_add(comp_len, &zram->stats.compr_data_size);
out:
/*
@@ -2136,6 +2140,8 @@ module_exit(zram_exit);
module_param(num_devices, uint, 0);
MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
+module_param_string(backend, backend_par_buf, BACKEND_PAR_BUF_SIZE, S_IRUGO);
+MODULE_PARM_DESC(backend, "Compression storage (backend) name");
MODULE_LICENSE("Dual BSD/GPL");
MODULE_AUTHOR("Nitin Gupta <[email protected]>");
diff --git a/drivers/block/zram/zram_drv.h b/drivers/block/zram/zram_drv.h
index f2fd46daa760..f4f51c6489ba 100644
--- a/drivers/block/zram/zram_drv.h
+++ b/drivers/block/zram/zram_drv.h
@@ -16,7 +16,7 @@
#define _ZRAM_DRV_H_
#include <linux/rwsem.h>
-#include <linux/zsmalloc.h>
+#include <linux/zpool.h>
#include <linux/crypto.h>
#include "zcomp.h"
@@ -91,7 +91,7 @@ struct zram_stats {
struct zram {
struct zram_table_entry *table;
- struct zs_pool *mem_pool;
+ struct zpool *mem_pool;
struct zcomp *comp;
struct gendisk *disk;
/* Prevent concurrent execution of device init */
--
2.20.1
Hi,
On (10/10/19 23:04), Vitaly Wool wrote:
[..]
> The coming patchset is a new take on the old issue: ZRAM can
> currently be used only with zsmalloc even though this may not
> be the optimal combination for some configurations. The previous
> (unsuccessful) attempt dates back to 2015 [1] and is notable for
> the heated discussions it has caused.
Oh, right, I do recall it.
> The patchset in [1] had basically the only goal of enabling
> ZRAM/zbud combo which had a very narrow use case. Things have
> changed substantially since then, and now, with z3fold used
> widely as a zswap backend, I, as the z3fold maintainer, am
> getting requests to re-interate on making it possible to use
> ZRAM with any zpool-compatible backend, first of all z3fold.
A quick question, what are the technical reasons to prefer
allocator X over zsmalloc? Some data would help, I guess.
> The preliminary results for this work have been delivered at
> Linux Plumbers this year [2]. The talk at LPC, though having
> attracted limited interest, ended in a consensus to continue
> the work and pursue the goal of decoupling ZRAM from zsmalloc.
[..]
> [1] https://lkml.org/lkml/2015/9/14/356
I need to re-read it, thanks for the link. IIRC, but maybe
I'm wrong, one of the things Minchan was not happy with was
increased maintenance cost. So, perhaps, this also should
be discuss/addressed (and maybe even in the first place).
-ss
On (10/10/19 23:11), Vitaly Wool wrote:
[..]
> +static unsigned long zs_zpool_get_compacted(void *pool)
> +{
> + struct zs_pool_stats stats;
> +
> + zs_pool_stats(pool, &stats);
> + return stats.pages_compacted;
> +}
So zs_pool_stats() can become static?
-ss
On (10/10/19 23:20), Vitaly Wool wrote:
[..]
> static const char *default_compressor = "lzo-rle";
>
> +#define BACKEND_PAR_BUF_SIZE 32
> +static char backend_par_buf[BACKEND_PAR_BUF_SIZE];
We can have multiple zram devices (zram0 .. zramN), I guess it
would make sense not to force all devices to use one particular
allocator (e.g. see comp_algorithm_store()).
If the motivation for the patch set is that zsmalloc does not
perform equally well for various data access patterns, then the
same is true for any other allocator. Thus, I think, we need to
have a per-device 'allocator' knob.
-ss
Hi Sergey,
On Mon, Oct 14, 2019 at 12:35 PM Sergey Senozhatsky
<[email protected]> wrote:
>
> Hi,
>
> On (10/10/19 23:04), Vitaly Wool wrote:
> [..]
> > The coming patchset is a new take on the old issue: ZRAM can
> > currently be used only with zsmalloc even though this may not
> > be the optimal combination for some configurations. The previous
> > (unsuccessful) attempt dates back to 2015 [1] and is notable for
> > the heated discussions it has caused.
>
> Oh, right, I do recall it.
>
> > The patchset in [1] had basically the only goal of enabling
> > ZRAM/zbud combo which had a very narrow use case. Things have
> > changed substantially since then, and now, with z3fold used
> > widely as a zswap backend, I, as the z3fold maintainer, am
> > getting requests to re-interate on making it possible to use
> > ZRAM with any zpool-compatible backend, first of all z3fold.
>
> A quick question, what are the technical reasons to prefer
> allocator X over zsmalloc? Some data would help, I guess.
For z3fold, the data can be found here:
https://elinux.org/images/d/d3/Z3fold.pdf.
For zbud (which is also of interest), imagine a low-end platform with
a simplistic HW compressor that doesn't give really high ratio. We
still want to be able to use ZRAM (not necessarily as a swap
partition, but rather for /home and /var) but we absolutely don't need
zsmalloc's complexity. zbud is a perfect match here (provided that it
can cope with PAGE_SIZE pages, yes, but it's a small patch to make
that work) since it's unlikely that we squeeze more than 2 compressed
pages per page with that HW compressor anyway.
> > The preliminary results for this work have been delivered at
> > Linux Plumbers this year [2]. The talk at LPC, though having
> > attracted limited interest, ended in a consensus to continue
> > the work and pursue the goal of decoupling ZRAM from zsmalloc.
>
> [..]
>
> > [1] https://lkml.org/lkml/2015/9/14/356
>
> I need to re-read it, thanks for the link. IIRC, but maybe
> I'm wrong, one of the things Minchan was not happy with was
> increased maintenance cost. So, perhaps, this also should
> be discuss/addressed (and maybe even in the first place).
I have hard time seeing how maintenance cost is increased here :)
~Vitaly
On Mon, Oct 14, 2019 at 12:49 PM Sergey Senozhatsky
<[email protected]> wrote:
>
> On (10/10/19 23:20), Vitaly Wool wrote:
> [..]
> > static const char *default_compressor = "lzo-rle";
> >
> > +#define BACKEND_PAR_BUF_SIZE 32
> > +static char backend_par_buf[BACKEND_PAR_BUF_SIZE];
>
> We can have multiple zram devices (zram0 .. zramN), I guess it
> would make sense not to force all devices to use one particular
> allocator (e.g. see comp_algorithm_store()).
>
> If the motivation for the patch set is that zsmalloc does not
> perform equally well for various data access patterns, then the
> same is true for any other allocator. Thus, I think, we need to
> have a per-device 'allocator' knob.
We were thinking here in per-SoC terms basically, but this is a valid
point. Since zram has a well-established sysfs per-device
configuration interface, backend choice better be moved there. Agree?
~Vitaly
On Thu, Oct 10, 2019 at 11:04:14PM +0300, Vitaly Wool wrote:
> The coming patchset is a new take on the old issue: ZRAM can currently be used only with zsmalloc even though this may not be the optimal combination for some configurations. The previous (unsuccessful) attempt dates back to 2015 [1] and is notable for the heated discussions it has caused.
>
> The patchset in [1] had basically the only goal of enabling ZRAM/zbud combo which had a very narrow use case. Things have changed substantially since then, and now, with z3fold used widely as a zswap backend, I, as the z3fold maintainer, am getting requests to re-interate on making it possible to use ZRAM with any zpool-compatible backend, first of all z3fold.
>
> The preliminary results for this work have been delivered at Linux Plumbers this year [2]. The talk at LPC, though having attracted limited interest, ended in a consensus to continue the work and pursue the goal of decoupling ZRAM from zsmalloc.
>
> The current patchset has been stress tested on arm64 and x86_64 devices, including the Dell laptop I'm writing this message on now, not to mention several QEmu confugirations.
>
> [1] https://lkml.org/lkml/2015/9/14/356
> [2] https://linuxplumbersconf.org/event/4/contributions/551/
Please describe what's the usecase in real world, what's the benefit zsmalloc
cannot fulfill by desgin and how it's significant.
I really don't want to make fragmentaion of allocator so we should really see
how zsmalloc cannot achieve things if you are claiming.
Please tell us how to test it so that we could investigate what's the root
cause.
On (10/14/19 13:52), Vitaly Wool wrote:
> On Mon, Oct 14, 2019 at 12:49 PM Sergey Senozhatsky
> <[email protected]> wrote:
> >
> > On (10/10/19 23:20), Vitaly Wool wrote:
> > [..]
> > > static const char *default_compressor = "lzo-rle";
> > >
> > > +#define BACKEND_PAR_BUF_SIZE 32
> > > +static char backend_par_buf[BACKEND_PAR_BUF_SIZE];
> >
> > We can have multiple zram devices (zram0 .. zramN), I guess it
> > would make sense not to force all devices to use one particular
> > allocator (e.g. see comp_algorithm_store()).
> >
> > If the motivation for the patch set is that zsmalloc does not
> > perform equally well for various data access patterns, then the
> > same is true for any other allocator. Thus, I think, we need to
> > have a per-device 'allocator' knob.
>
> We were thinking here in per-SoC terms basically, but this is a valid
> point. Since zram has a well-established sysfs per-device
> configuration interface, backend choice better be moved there. Agree?
Yup, sysfs per-device knob.
// Given that Minchan is OK with the patch set.
-ss
Hi Minchan,
On Mon, Oct 14, 2019 at 6:41 PM Minchan Kim <[email protected]> wrote:
>
> On Thu, Oct 10, 2019 at 11:04:14PM +0300, Vitaly Wool wrote:
> > The coming patchset is a new take on the old issue: ZRAM can currently be used only with zsmalloc even though this may not be the optimal combination for some configurations. The previous (unsuccessful) attempt dates back to 2015 [1] and is notable for the heated discussions it has caused.
> >
> > The patchset in [1] had basically the only goal of enabling ZRAM/zbud combo which had a very narrow use case. Things have changed substantially since then, and now, with z3fold used widely as a zswap backend, I, as the z3fold maintainer, am getting requests to re-interate on making it possible to use ZRAM with any zpool-compatible backend, first of all z3fold.
> >
> > The preliminary results for this work have been delivered at Linux Plumbers this year [2]. The talk at LPC, though having attracted limited interest, ended in a consensus to continue the work and pursue the goal of decoupling ZRAM from zsmalloc.
> >
> > The current patchset has been stress tested on arm64 and x86_64 devices, including the Dell laptop I'm writing this message on now, not to mention several QEmu confugirations.
> >
> > [1] https://lkml.org/lkml/2015/9/14/356
> > [2] https://linuxplumbersconf.org/event/4/contributions/551/
>
> Please describe what's the usecase in real world, what's the benefit zsmalloc
> cannot fulfill by desgin and how it's significant.
I'm not entirely sure how to interpret the phrase "the benefit
zsmalloc cannot fulfill by design" but let me explain.
First, there are multi multi core systems where z3fold can provide
better throughput.
Then, there are low end systems with hardware
compression/decompression support which don't need zsmalloc
sophistication and would rather use zbud with ZRAM because the
compression ratio is relatively low.
Finally, there are MMU-less systems targeting IOT and still running
Linux and having a compressed RAM disk is something that would help
these systems operate in a better way (for the benefit of the overall
Linux ecosystem, if you care about that, of course; well, some people
do).
> I really don't want to make fragmentaion of allocator so we should really see
> how zsmalloc cannot achieve things if you are claiming.
I have to say that this point is completely bogus. We do not create
fragmentation by using a better defined and standardized API. In fact,
we aim to increase the number of use cases and test coverage for ZRAM.
With that said, I have hard time seeing how zsmalloc can operate on a
MMU-less system.
> Please tell us how to test it so that we could investigate what's the root
> cause.
I gather you haven't read neither the LPC documents nor my
conversation with Sergey re: these changes, because if you did you
wouldn't have had the type of questions you're asking. Please also see
above.
I feel a bit awkward explaining basic things to you but there may not
be other "root cause" than applicability issue. zsmalloc is a great
allocator but it's not universal and has its limitations. The
(potential) scope for ZRAM is wider than zsmalloc can provide. We are
*helping* _you_ to extend this scope "in real world" (c) and you come
up with bogus objections. Why?
Best regards,
Vitaly
On Tue, Oct 15, 2019 at 09:39:35AM +0200, Vitaly Wool wrote:
> Hi Minchan,
>
> On Mon, Oct 14, 2019 at 6:41 PM Minchan Kim <[email protected]> wrote:
> >
> > On Thu, Oct 10, 2019 at 11:04:14PM +0300, Vitaly Wool wrote:
> > > The coming patchset is a new take on the old issue: ZRAM can currently be used only with zsmalloc even though this may not be the optimal combination for some configurations. The previous (unsuccessful) attempt dates back to 2015 [1] and is notable for the heated discussions it has caused.
> > >
> > > The patchset in [1] had basically the only goal of enabling ZRAM/zbud combo which had a very narrow use case. Things have changed substantially since then, and now, with z3fold used widely as a zswap backend, I, as the z3fold maintainer, am getting requests to re-interate on making it possible to use ZRAM with any zpool-compatible backend, first of all z3fold.
> > >
> > > The preliminary results for this work have been delivered at Linux Plumbers this year [2]. The talk at LPC, though having attracted limited interest, ended in a consensus to continue the work and pursue the goal of decoupling ZRAM from zsmalloc.
> > >
> > > The current patchset has been stress tested on arm64 and x86_64 devices, including the Dell laptop I'm writing this message on now, not to mention several QEmu confugirations.
> > >
> > > [1] https://lkml.org/lkml/2015/9/14/356
> > > [2] https://linuxplumbersconf.org/event/4/contributions/551/
> >
> > Please describe what's the usecase in real world, what's the benefit zsmalloc
> > cannot fulfill by desgin and how it's significant.
>
> I'm not entirely sure how to interpret the phrase "the benefit
> zsmalloc cannot fulfill by design" but let me explain.
> First, there are multi multi core systems where z3fold can provide
> better throughput.
Please include number in the description with workload.
> Then, there are low end systems with hardware
> compression/decompression support which don't need zsmalloc
> sophistication and would rather use zbud with ZRAM because the
> compression ratio is relatively low.
I couldn't imagine how it's bad with zsmalloc. Could you be more
specific?
> Finally, there are MMU-less systems targeting IOT and still running
> Linux and having a compressed RAM disk is something that would help
> these systems operate in a better way (for the benefit of the overall
> Linux ecosystem, if you care about that, of course; well, some people
> do).
Could you write down what's the problem to use zsmalloc for MMU-less
system? Maybe, it would be important point rather other performance
argument since other functions's overheads in the callpath are already
rather big.
>
> > I really don't want to make fragmentaion of allocator so we should really see
> > how zsmalloc cannot achieve things if you are claiming.
>
> I have to say that this point is completely bogus. We do not create
> fragmentation by using a better defined and standardized API. In fact,
> we aim to increase the number of use cases and test coverage for ZRAM.
> With that said, I have hard time seeing how zsmalloc can operate on a
> MMU-less system.
>
> > Please tell us how to test it so that we could investigate what's the root
> > cause.
>
> I gather you haven't read neither the LPC documents nor my
> conversation with Sergey re: these changes, because if you did you
> wouldn't have had the type of questions you're asking. Please also see
> above.
Please include your claims in the description rather than attaching
file. That's the usualy way how we work because it could make easier to
discuss by inline.
>
> I feel a bit awkward explaining basic things to you but there may not
> be other "root cause" than applicability issue. zsmalloc is a great
> allocator but it's not universal and has its limitations. The
> (potential) scope for ZRAM is wider than zsmalloc can provide. We are
> *helping* _you_ to extend this scope "in real world" (c) and you come
> up with bogus objections. Why?
Please add more detail to convince so we need to think over why zsmalloc
cannot be improved for the usecase.
On Thu, Oct 10, 2019 at 4:09 PM Vitaly Wool <[email protected]> wrote:
>
> This patch adds the following functions to the zpool API:
> - zpool_compact()
> - zpool_get_num_compacted()
> - zpool_huge_class_size()
>
> The first one triggers compaction for the underlying allocator, the
> second retrieves the number of pages migrated due to compaction for
> the whole time of this pool's existence and the third one returns
> the huge class size.
>
> This API extension is done to align zpool API with zsmalloc API.
>
> Signed-off-by: Vitaly Wool <[email protected]>
Seems reasonable to me.
Reviewed-by: Dan Streetman <[email protected]>
> ---
> include/linux/zpool.h | 14 +++++++++++++-
> mm/zpool.c | 36 ++++++++++++++++++++++++++++++++++++
> 2 files changed, 49 insertions(+), 1 deletion(-)
>
> diff --git a/include/linux/zpool.h b/include/linux/zpool.h
> index 51bf43076165..31f0c1360569 100644
> --- a/include/linux/zpool.h
> +++ b/include/linux/zpool.h
> @@ -61,8 +61,13 @@ void *zpool_map_handle(struct zpool *pool, unsigned long handle,
>
> void zpool_unmap_handle(struct zpool *pool, unsigned long handle);
>
> +unsigned long zpool_compact(struct zpool *pool);
> +
> +unsigned long zpool_get_num_compacted(struct zpool *pool);
> +
> u64 zpool_get_total_size(struct zpool *pool);
>
> +size_t zpool_huge_class_size(struct zpool *zpool);
>
> /**
> * struct zpool_driver - driver implementation for zpool
> @@ -75,7 +80,10 @@ u64 zpool_get_total_size(struct zpool *pool);
> * @shrink: shrink the pool.
> * @map: map a handle.
> * @unmap: unmap a handle.
> - * @total_size: get total size of a pool.
> + * @compact: try to run compaction over a pool
> + * @get_num_compacted: get amount of compacted pages for a pool
> + * @total_size: get total size of a pool
> + * @huge_class_size: huge class threshold for pool pages.
> *
> * This is created by a zpool implementation and registered
> * with zpool.
> @@ -104,7 +112,11 @@ struct zpool_driver {
> enum zpool_mapmode mm);
> void (*unmap)(void *pool, unsigned long handle);
>
> + unsigned long (*compact)(void *pool);
> + unsigned long (*get_num_compacted)(void *pool);
> +
> u64 (*total_size)(void *pool);
> + size_t (*huge_class_size)(void *pool);
> };
>
> void zpool_register_driver(struct zpool_driver *driver);
> diff --git a/mm/zpool.c b/mm/zpool.c
> index 863669212070..55e69213c2eb 100644
> --- a/mm/zpool.c
> +++ b/mm/zpool.c
> @@ -362,6 +362,30 @@ void zpool_unmap_handle(struct zpool *zpool, unsigned long handle)
> zpool->driver->unmap(zpool->pool, handle);
> }
>
> + /**
> + * zpool_compact() - try to run compaction over zpool
> + * @pool The zpool to compact
> + *
> + * Returns: the number of migrated pages
> + */
> +unsigned long zpool_compact(struct zpool *zpool)
> +{
> + return zpool->driver->compact ? zpool->driver->compact(zpool->pool) : 0;
> +}
> +
> +
> +/**
> + * zpool_get_num_compacted() - get the number of migrated/compacted pages
> + * @pool The zpool to get compaction statistic for
> + *
> + * Returns: the total number of migrated pages for the pool
> + */
> +unsigned long zpool_get_num_compacted(struct zpool *zpool)
> +{
> + return zpool->driver->get_num_compacted ?
> + zpool->driver->get_num_compacted(zpool->pool) : 0;
> +}
> +
> /**
> * zpool_get_total_size() - The total size of the pool
> * @zpool: The zpool to check
> @@ -375,6 +399,18 @@ u64 zpool_get_total_size(struct zpool *zpool)
> return zpool->driver->total_size(zpool->pool);
> }
>
> +/**
> + * zpool_huge_class_size() - get size for the "huge" class
> + * @pool The zpool to check
> + *
> + * Returns: size of the huge class
> + */
> +size_t zpool_huge_class_size(struct zpool *zpool)
> +{
> + return zpool->driver->huge_class_size ?
> + zpool->driver->huge_class_size(zpool->pool) : 0;
> +}
> +
> /**
> * zpool_evictable() - Test if zpool is potentially evictable
> * @zpool: The zpool to test
> --
> 2.20.1
On Tue, Oct 15, 2019 at 10:00 PM Minchan Kim <[email protected]> wrote:
>
> On Tue, Oct 15, 2019 at 09:39:35AM +0200, Vitaly Wool wrote:
> > Hi Minchan,
> >
> > On Mon, Oct 14, 2019 at 6:41 PM Minchan Kim <[email protected]> wrote:
> > >
> > > On Thu, Oct 10, 2019 at 11:04:14PM +0300, Vitaly Wool wrote:
> > > > The coming patchset is a new take on the old issue: ZRAM can currently be used only with zsmalloc even though this may not be the optimal combination for some configurations. The previous (unsuccessful) attempt dates back to 2015 [1] and is notable for the heated discussions it has caused.
> > > >
> > > > The patchset in [1] had basically the only goal of enabling ZRAM/zbud combo which had a very narrow use case. Things have changed substantially since then, and now, with z3fold used widely as a zswap backend, I, as the z3fold maintainer, am getting requests to re-interate on making it possible to use ZRAM with any zpool-compatible backend, first of all z3fold.
> > > >
> > > > The preliminary results for this work have been delivered at Linux Plumbers this year [2]. The talk at LPC, though having attracted limited interest, ended in a consensus to continue the work and pursue the goal of decoupling ZRAM from zsmalloc.
> > > >
> > > > The current patchset has been stress tested on arm64 and x86_64 devices, including the Dell laptop I'm writing this message on now, not to mention several QEmu confugirations.
> > > >
> > > > [1] https://lkml.org/lkml/2015/9/14/356
> > > > [2] https://linuxplumbersconf.org/event/4/contributions/551/
> > >
> > > Please describe what's the usecase in real world, what's the benefit zsmalloc
> > > cannot fulfill by desgin and how it's significant.
> >
> > I'm not entirely sure how to interpret the phrase "the benefit
> > zsmalloc cannot fulfill by design" but let me explain.
> > First, there are multi multi core systems where z3fold can provide
> > better throughput.
>
> Please include number in the description with workload.
Sure. So on an HMP 8-core ARM64 system with ZRAM, we run the following command:
fio --bs=4k --randrepeat=1 --randseed=100 --refill_buffers \
--buffer_compress_percentage=50 --scramble_buffers=1 \
--direct=1 --loops=15 --numjobs=4 --filename=/dev/block/zram0 \
--name=seq-write --rw=write --stonewall --name=seq-read \
--rw=read --stonewall --name=seq-readwrite --rw=rw --stonewall \
--name=rand-readwrite --rw=randrw --stonewall
The results are the following:
zsmalloc:
Run status group 0 (all jobs):
WRITE: io=61440MB, aggrb=1680.4MB/s, minb=430167KB/s,
maxb=440590KB/s, mint=35699msec, maxt=36564msec
Run status group 1 (all jobs):
READ: io=61440MB, aggrb=1620.4MB/s, minb=414817KB/s,
maxb=414850KB/s, mint=37914msec, maxt=37917msec
Run status group 2 (all jobs):
READ: io=30615MB, aggrb=897979KB/s, minb=224494KB/s,
maxb=228161KB/s, mint=34351msec, maxt=34912msec
WRITE: io=30825MB, aggrb=904110KB/s, minb=226027KB/s,
maxb=229718KB/s, mint=34351msec, maxt=34912msec
Run status group 3 (all jobs):
READ: io=30615MB, aggrb=772002KB/s, minb=193000KB/s,
maxb=193010KB/s, mint=40607msec, maxt=40609msec
WRITE: io=30825MB, aggrb=777273KB/s, minb=194318KB/s,
maxb=194327KB/s, mint=40607msec, maxt=40609msec
z3fold:
Run status group 0 (all jobs):
WRITE: io=61440MB, aggrb=1224.8MB/s, minb=313525KB/s,
maxb=329941KB/s, mint=47671msec, maxt=50167msec
Run status group 1 (all jobs):
READ: io=61440MB, aggrb=3119.3MB/s, minb=798529KB/s,
maxb=862883KB/s, mint=18228msec, maxt=19697msec
Run status group 2 (all jobs):
READ: io=30615MB, aggrb=937283KB/s, minb=234320KB/s,
maxb=234334KB/s, mint=33446msec, maxt=33448msec
WRITE: io=30825MB, aggrb=943682KB/s, minb=235920KB/s,
maxb=235934KB/s, mint=33446msec, maxt=33448msec
Run status group 3 (all jobs):
READ: io=30615MB, aggrb=829591KB/s, minb=207397KB/s,
maxb=210285KB/s, mint=37271msec, maxt=37790msec
WRITE: io=30825MB, aggrb=835255KB/s, minb=208813KB/s,
maxb=211721KB/s, mint=37271msec, maxt=37790msec
So, z3fold is faster everywhere (including being *two* times faster on
read) except for sequential write which is the least important use
case in real world.
> > Then, there are low end systems with hardware
> > compression/decompression support which don't need zsmalloc
> > sophistication and would rather use zbud with ZRAM because the
> > compression ratio is relatively low.
>
> I couldn't imagine how it's bad with zsmalloc. Could you be more
> specific?
> > Finally, there are MMU-less systems targeting IOT and still running
> > Linux and having a compressed RAM disk is something that would help
> > these systems operate in a better way (for the benefit of the overall
> > Linux ecosystem, if you care about that, of course; well, some people
> > do).
>
> Could you write down what's the problem to use zsmalloc for MMU-less
> system? Maybe, it would be important point rather other performance
> argument since other functions's overheads in the callpath are already
> rather big.
Well, I assume you had the reasons to make zsmalloc depend on MMU in Kconfig:
...
config ZSMALLOC
tristate "Memory allocator for compressed pages"
depends on MMU
help
...
But even disregarding that, let's compare ZRAM/zbud and ZRAM/zsmalloc
performance and memory these two consume on a relatively low end
2-core ARM.
Command:
fio --bs=4k --randrepeat=1 --randseed=100 --refill_buffers
--scramble_buffers=1 \
--direct=1 --loops=15 --numjobs=2 --filename=/dev/block/zram0 \
--name=seq-write --rw=write --stonewall --name=seq-read --rw=read \
--stonewall --name=seq-readwrite --rw=rw --stonewall
--name=rand-readwrite \
--rw=randrw --stonewall
zsmalloc:
Run status group 0 (all jobs):
WRITE: io=30720MB, aggrb=374763KB/s, minb=187381KB/s,
maxb=188389KB/s, mint=83490msec, maxt=83939msec
Run status group 1 (all jobs):
READ: io=30720MB, aggrb=964000KB/s, minb=482000KB/s,
maxb=482015KB/s, mint=32631msec, maxt=32632msec
Run status group 2 (all jobs):
READ: io=15308MB, aggrb=431263KB/s, minb=215631KB/s,
maxb=215898KB/s, mint=36302msec, maxt=36347msec
WRITE: io=15412MB, aggrb=434207KB/s, minb=217103KB/s,
maxb=217373KB/s, mint=36302msec, maxt=36347msec
Run status group 3 (all jobs):
READ: io=15308MB, aggrb=327328KB/s, minb=163664KB/s,
maxb=163667KB/s, mint=47887msec, maxt=47888msec
WRITE: io=15412MB, aggrb=329563KB/s, minb=164781KB/s,
maxb=164785KB/s, mint=47887msec, maxt=47888msec
zbud:
Run status group 0 (all jobs):
WRITE: io=30720MB, aggrb=735980KB/s, minb=367990KB/s,
maxb=373079KB/s, mint=42159msec, maxt=42742msec
Run status group 1 (all jobs):
READ: io=30720MB, aggrb=927915KB/s, minb=463957KB/s,
maxb=463999KB/s, mint=33898msec, maxt=33901msec
Run status group 2 (all jobs):
READ: io=15308MB, aggrb=403467KB/s, minb=201733KB/s,
maxb=202051KB/s, mint=38790msec, maxt=38851msec
WRITE: io=15412MB, aggrb=406222KB/s, minb=203111KB/s,
maxb=203430KB/s, mint=38790msec, maxt=38851msec
Run status group 3 (all jobs):
READ: io=15308MB, aggrb=334967KB/s, minb=167483KB/s,
maxb=167487KB/s, mint=46795msec, maxt=46796msec
WRITE: io=15412MB, aggrb=337254KB/s, minb=168627KB/s,
maxb=168630KB/s, mint=46795msec, maxt=46796msec
Pretty equal except for sequential write which is twice as good with zbud.
Now to the fun part.
zsmalloc:
0 .text 00002908 0000000000000000 0000000000000000 00000040 2**2
CONTENTS, ALLOC, LOAD, RELOC, READONLY, CODE
zbud:
0 .text 0000072c 0000000000000000 0000000000000000 00000040 2**2
CONTENTS, ALLOC, LOAD, RELOC, READONLY, CODE
And this does not cover dynamic memory allocation overhead which is
higher for zsmalloc. So once again, given that the compression ratio
is low (e. g. a simple HW accelerator is used), what would most
unbiased people prefer to use in this case?
> > > I really don't want to make fragmentaion of allocator so we should really see
> > > how zsmalloc cannot achieve things if you are claiming.
> >
> > I have to say that this point is completely bogus. We do not create
> > fragmentation by using a better defined and standardized API. In fact,
> > we aim to increase the number of use cases and test coverage for ZRAM.
> > With that said, I have hard time seeing how zsmalloc can operate on a
> > MMU-less system.
> >
> > > Please tell us how to test it so that we could investigate what's the root
> > > cause.
> >
> > I gather you haven't read neither the LPC documents nor my
> > conversation with Sergey re: these changes, because if you did you
> > wouldn't have had the type of questions you're asking. Please also see
> > above.
>
> Please include your claims in the description rather than attaching
> file. That's the usualy way how we work because it could make easier to
> discuss by inline.
Did I attach something? I don't quite recall that. I posted links to
previous discussions and conference materials, each for a reason.
> >
> > I feel a bit awkward explaining basic things to you but there may not
> > be other "root cause" than applicability issue. zsmalloc is a great
> > allocator but it's not universal and has its limitations. The
> > (potential) scope for ZRAM is wider than zsmalloc can provide. We are
> > *helping* _you_ to extend this scope "in real world" (c) and you come
> > up with bogus objections. Why?
>
> Please add more detail to convince so we need to think over why zsmalloc
> cannot be improved for the usecase.
This approach is wrong. zsmalloc is good enough and covers a lot of
use cases but there are still some where it doesn't work that well by
design. E. g. on an XIP system we do care about the code size since
it's stored uncompressed but still want to use ZRAM. Why would we want
to waste almost 10K just on zsmalloc code if the counterpart (zbud in
that case) works better?
Best regards,
Vitaly
On Mon, Oct 21, 2019 at 04:21:21PM +0200, Vitaly Wool wrote:
> On Tue, Oct 15, 2019 at 10:00 PM Minchan Kim <[email protected]> wrote:
> >
> > On Tue, Oct 15, 2019 at 09:39:35AM +0200, Vitaly Wool wrote:
> > > Hi Minchan,
> > >
> > > On Mon, Oct 14, 2019 at 6:41 PM Minchan Kim <[email protected]> wrote:
> > > >
> > > > On Thu, Oct 10, 2019 at 11:04:14PM +0300, Vitaly Wool wrote:
> > > > > The coming patchset is a new take on the old issue: ZRAM can currently be used only with zsmalloc even though this may not be the optimal combination for some configurations. The previous (unsuccessful) attempt dates back to 2015 [1] and is notable for the heated discussions it has caused.
> > > > >
> > > > > The patchset in [1] had basically the only goal of enabling ZRAM/zbud combo which had a very narrow use case. Things have changed substantially since then, and now, with z3fold used widely as a zswap backend, I, as the z3fold maintainer, am getting requests to re-interate on making it possible to use ZRAM with any zpool-compatible backend, first of all z3fold.
> > > > >
> > > > > The preliminary results for this work have been delivered at Linux Plumbers this year [2]. The talk at LPC, though having attracted limited interest, ended in a consensus to continue the work and pursue the goal of decoupling ZRAM from zsmalloc.
> > > > >
> > > > > The current patchset has been stress tested on arm64 and x86_64 devices, including the Dell laptop I'm writing this message on now, not to mention several QEmu confugirations.
> > > > >
> > > > > [1] https://lkml.org/lkml/2015/9/14/356
> > > > > [2] https://linuxplumbersconf.org/event/4/contributions/551/
> > > >
> > > > Please describe what's the usecase in real world, what's the benefit zsmalloc
> > > > cannot fulfill by desgin and how it's significant.
> > >
> > > I'm not entirely sure how to interpret the phrase "the benefit
> > > zsmalloc cannot fulfill by design" but let me explain.
> > > First, there are multi multi core systems where z3fold can provide
> > > better throughput.
> >
> > Please include number in the description with workload.
>
> Sure. So on an HMP 8-core ARM64 system with ZRAM, we run the following command:
> fio --bs=4k --randrepeat=1 --randseed=100 --refill_buffers \
> --buffer_compress_percentage=50 --scramble_buffers=1 \
> --direct=1 --loops=15 --numjobs=4 --filename=/dev/block/zram0 \
> --name=seq-write --rw=write --stonewall --name=seq-read \
> --rw=read --stonewall --name=seq-readwrite --rw=rw --stonewall \
> --name=rand-readwrite --rw=randrw --stonewall
>
> The results are the following:
>
> zsmalloc:
> Run status group 0 (all jobs):
> WRITE: io=61440MB, aggrb=1680.4MB/s, minb=430167KB/s,
> maxb=440590KB/s, mint=35699msec, maxt=36564msec
>
> Run status group 1 (all jobs):
> READ: io=61440MB, aggrb=1620.4MB/s, minb=414817KB/s,
> maxb=414850KB/s, mint=37914msec, maxt=37917msec
>
> Run status group 2 (all jobs):
> READ: io=30615MB, aggrb=897979KB/s, minb=224494KB/s,
> maxb=228161KB/s, mint=34351msec, maxt=34912msec
> WRITE: io=30825MB, aggrb=904110KB/s, minb=226027KB/s,
> maxb=229718KB/s, mint=34351msec, maxt=34912msec
>
> Run status group 3 (all jobs):
> READ: io=30615MB, aggrb=772002KB/s, minb=193000KB/s,
> maxb=193010KB/s, mint=40607msec, maxt=40609msec
> WRITE: io=30825MB, aggrb=777273KB/s, minb=194318KB/s,
> maxb=194327KB/s, mint=40607msec, maxt=40609msec
>
> z3fold:
> Run status group 0 (all jobs):
> WRITE: io=61440MB, aggrb=1224.8MB/s, minb=313525KB/s,
> maxb=329941KB/s, mint=47671msec, maxt=50167msec
>
> Run status group 1 (all jobs):
> READ: io=61440MB, aggrb=3119.3MB/s, minb=798529KB/s,
> maxb=862883KB/s, mint=18228msec, maxt=19697msec
>
> Run status group 2 (all jobs):
> READ: io=30615MB, aggrb=937283KB/s, minb=234320KB/s,
> maxb=234334KB/s, mint=33446msec, maxt=33448msec
> WRITE: io=30825MB, aggrb=943682KB/s, minb=235920KB/s,
> maxb=235934KB/s, mint=33446msec, maxt=33448msec
>
> Run status group 3 (all jobs):
> READ: io=30615MB, aggrb=829591KB/s, minb=207397KB/s,
> maxb=210285KB/s, mint=37271msec, maxt=37790msec
> WRITE: io=30825MB, aggrb=835255KB/s, minb=208813KB/s,
> maxb=211721KB/s, mint=37271msec, maxt=37790msec
>
> So, z3fold is faster everywhere (including being *two* times faster on
> read) except for sequential write which is the least important use
> case in real world.
No. write is also important because it affects reclaim speed.
I ran fio on x86 with various compression sizes.
left is zsmalloc. right is z3fold
The operation order is
seq-write
rand-write
seq-read
rand-read
mixed-seq
mixed-rand
trim
mem_used - byte unit
Last column mem_used is to indicate how many allocator used the memory
to store compressed page
1) compression ratio 75
WRITE 2535 WRITE 1928
WRITE 2425 WRITE 1886
READ 6211 READ 5731
READ 6339 READ 6182
READ 1791 READ 1592
WRITE 1790 WRITE 1591
READ 1704 READ 1493
WRITE 1699 WRITE 1489
WRITE 984 WRITE 974
TRIM 984 TRIM 974
mem_used 29986816 mem_used 61239296
For every operation, zsmalloc is faster than z3fold.
Even, it used the 1/2 memory compared to z3fold.
2) compression ratio 66
WRITE 2125 WRITE 1258
WRITE 2107 WRITE 1233
READ 5714 READ 5793
READ 5948 READ 6065
READ 1667 READ 1248
WRITE 1666 WRITE 1247
READ 1521 READ 1218
WRITE 1517 WRITE 1215
WRITE 943 WRITE 870
TRIM 943 TRIM 870
mem_used 38158336 mem_used 76779520
For only read operation, z3fold is a bit faster than zsmalloc about 2%.
However, look at other operations which zsmalloc is much faster.
Even, look at used memory.
3) compression ratio 50
WRITE 2051 WRITE 1109
WRITE 2029 WRITE 1087
READ 5366 READ 6364
READ 5575 READ 5785
READ 1497 READ 1121
WRITE 1496 WRITE 1121
READ 1432 READ 1065
WRITE 1428 WRITE 1062
WRITE 930 WRITE 838
TRIM 930 TRIM 838
mem_used 59932672 mem_used 104873984
sequential read on z3fold is faster about 15%. However, look at other
operations and used memory. zsmalloc is better.
Why zsmalloc is slow for 50% compression ratio is it needs page copy
for every read operation since compressed objects cross over page boundary.
However, I don't think it's real workload because compressed ratio will
spread out into various sizes. Having said that, I could enhance zsmalloc
to avoid the copy operation. I will work on it.
4) compression ratio 33
WRITE 1945 WRITE 1239
WRITE 1869 WRITE 1222
READ 5319 READ 6206
READ 5416 READ 6645
READ 1480 READ 1188
WRITE 1479 WRITE 1188
READ 1403 READ 1114
WRITE 1399 WRITE 1110
WRITE 930 WRITE 793
TRIM 930 TRIM 793
mem_used 78667776 mem_used 104873984
5) compression ratio 25
WRITE 1862 WRITE 1080
WRITE 1840 WRITE 1052
READ 5260 READ 6240
READ 5540 READ 6359
READ 1445 READ 1040
WRITE 1444 WRITE 1039
READ 1354 READ 1006
WRITE 1350 WRITE 1003
WRITE 909 WRITE 775
TRIM 909 TRIM 775
mem_used 83902464 mem_used 104873984
If compress ratio is bad, zram read operation with zsmalloc could
be slower about 15% than z3fold because it needs additional memory
copy as I mentioned. However, it's still faster if compression ratio
ig greater than 50%, which is usual case(I believe that's why you
makes z3fold).
>
> > > Then, there are low end systems with hardware
> > > compression/decompression support which don't need zsmalloc
> > > sophistication and would rather use zbud with ZRAM because the
> > > compression ratio is relatively low.
> >
> > I couldn't imagine how it's bad with zsmalloc. Could you be more
> > specific?
>
>
> > > Finally, there are MMU-less systems targeting IOT and still running
> > > Linux and having a compressed RAM disk is something that would help
> > > these systems operate in a better way (for the benefit of the overall
> > > Linux ecosystem, if you care about that, of course; well, some people
> > > do).
> >
> > Could you write down what's the problem to use zsmalloc for MMU-less
> > system? Maybe, it would be important point rather other performance
> > argument since other functions's overheads in the callpath are already
> > rather big.
>
> Well, I assume you had the reasons to make zsmalloc depend on MMU in Kconfig:
> ...
> config ZSMALLOC
> tristate "Memory allocator for compressed pages"
> depends on MMU
> help
> ...
It's old piece left since zsmalloc used mapping API so I think we could
remove the dependency now. However, I want to know it's the only problem
to use zram in MMU-less system. IOW, if we could remove the zsmalloc MMU
dependency, it's ready to use zram on MMU-less system now?
>
> But even disregarding that, let's compare ZRAM/zbud and ZRAM/zsmalloc
> performance and memory these two consume on a relatively low end
> 2-core ARM.
> Command:
> fio --bs=4k --randrepeat=1 --randseed=100 --refill_buffers
> --scramble_buffers=1 \
> --direct=1 --loops=15 --numjobs=2 --filename=/dev/block/zram0 \
> --name=seq-write --rw=write --stonewall --name=seq-read --rw=read \
> --stonewall --name=seq-readwrite --rw=rw --stonewall
> --name=rand-readwrite \
> --rw=randrw --stonewall
>
> zsmalloc:
> Run status group 0 (all jobs):
> WRITE: io=30720MB, aggrb=374763KB/s, minb=187381KB/s,
> maxb=188389KB/s, mint=83490msec, maxt=83939msec
>
> Run status group 1 (all jobs):
> READ: io=30720MB, aggrb=964000KB/s, minb=482000KB/s,
> maxb=482015KB/s, mint=32631msec, maxt=32632msec
>
> Run status group 2 (all jobs):
> READ: io=15308MB, aggrb=431263KB/s, minb=215631KB/s,
> maxb=215898KB/s, mint=36302msec, maxt=36347msec
> WRITE: io=15412MB, aggrb=434207KB/s, minb=217103KB/s,
> maxb=217373KB/s, mint=36302msec, maxt=36347msec
>
> Run status group 3 (all jobs):
> READ: io=15308MB, aggrb=327328KB/s, minb=163664KB/s,
> maxb=163667KB/s, mint=47887msec, maxt=47888msec
> WRITE: io=15412MB, aggrb=329563KB/s, minb=164781KB/s,
> maxb=164785KB/s, mint=47887msec, maxt=47888msec
>
> zbud:
> Run status group 0 (all jobs):
> WRITE: io=30720MB, aggrb=735980KB/s, minb=367990KB/s,
> maxb=373079KB/s, mint=42159msec, maxt=42742msec
>
> Run status group 1 (all jobs):
> READ: io=30720MB, aggrb=927915KB/s, minb=463957KB/s,
> maxb=463999KB/s, mint=33898msec, maxt=33901msec
>
> Run status group 2 (all jobs):
> READ: io=15308MB, aggrb=403467KB/s, minb=201733KB/s,
> maxb=202051KB/s, mint=38790msec, maxt=38851msec
> WRITE: io=15412MB, aggrb=406222KB/s, minb=203111KB/s,
> maxb=203430KB/s, mint=38790msec, maxt=38851msec
>
> Run status group 3 (all jobs):
> READ: io=15308MB, aggrb=334967KB/s, minb=167483KB/s,
> maxb=167487KB/s, mint=46795msec, maxt=46796msec
> WRITE: io=15412MB, aggrb=337254KB/s, minb=168627KB/s,
> maxb=168630KB/s, mint=46795msec, maxt=46796msec
>
> Pretty equal except for sequential write which is twice as good with zbud.
Thanks for the testing. I also tried to test zbud with zram but failed because fio
submit incompressible pages to zram even though it specifiy compress ratio 100%
However, zbud doesn't support 4K page allocation so zram couldn't work on it
at this moment. I tried various fio versions as well as old but everything failed.
How did you test it successfully? Let me know your fio version.
I want to investigate what's the performance bottleneck beside page copy
so that I will optimize it.
>
> Now to the fun part.
> zsmalloc:
> 0 .text 00002908 0000000000000000 0000000000000000 00000040 2**2
> CONTENTS, ALLOC, LOAD, RELOC, READONLY, CODE
> zbud:
> 0 .text 0000072c 0000000000000000 0000000000000000 00000040 2**2
> CONTENTS, ALLOC, LOAD, RELOC, READONLY, CODE
>
> And this does not cover dynamic memory allocation overhead which is
> higher for zsmalloc. So once again, given that the compression ratio
> is low (e. g. a simple HW accelerator is used), what would most
> unbiased people prefer to use in this case?
Zsmalloc has more features than zbud. That's why you see the code size
difference. It was intentional because at that time most of users were
mobile phones, TV and other smart devices. They needed those features.
We could make those feature turned off at build time, which will improve
performance and reduce code size a lot. It would be no problem if the
user wanted to use zbud which is alredy lacking of those features.
>
> > > > I really don't want to make fragmentaion of allocator so we should really see
> > > > how zsmalloc cannot achieve things if you are claiming.
> > >
> > > I have to say that this point is completely bogus. We do not create
> > > fragmentation by using a better defined and standardized API. In fact,
> > > we aim to increase the number of use cases and test coverage for ZRAM.
> > > With that said, I have hard time seeing how zsmalloc can operate on a
> > > MMU-less system.
> > >
> > > > Please tell us how to test it so that we could investigate what's the root
> > > > cause.
> > >
> > > I gather you haven't read neither the LPC documents nor my
> > > conversation with Sergey re: these changes, because if you did you
> > > wouldn't have had the type of questions you're asking. Please also see
> > > above.
> >
> > Please include your claims in the description rather than attaching
> > file. That's the usualy way how we work because it could make easier to
> > discuss by inline.
>
> Did I attach something? I don't quite recall that. I posted links to
> previous discussions and conference materials, each for a reason.
>
> > >
> > > I feel a bit awkward explaining basic things to you but there may not
> > > be other "root cause" than applicability issue. zsmalloc is a great
> > > allocator but it's not universal and has its limitations. The
> > > (potential) scope for ZRAM is wider than zsmalloc can provide. We are
> > > *helping* _you_ to extend this scope "in real world" (c) and you come
> > > up with bogus objections. Why?
> >
> > Please add more detail to convince so we need to think over why zsmalloc
> > cannot be improved for the usecase.
>
> This approach is wrong. zsmalloc is good enough and covers a lot of
> use cases but there are still some where it doesn't work that well by
> design. E. g. on an XIP system we do care about the code size since
> it's stored uncompressed but still want to use ZRAM. Why would we want
> to waste almost 10K just on zsmalloc code if the counterpart (zbud in
> that case) works better?
As I mentiond, we could improve zsmalloc to reduce code size as well as
performance. I will work on it.
Hi Minchan,
On Wed, Oct 30, 2019 at 1:10 AM Minchan Kim <[email protected]> wrote:
<snip>
>
>
> I ran fio on x86 with various compression sizes.
>
> left is zsmalloc. right is z3fold
>
> The operation order is
> seq-write
> rand-write
> seq-read
> rand-read
> mixed-seq
> mixed-rand
> trim
> mem_used - byte unit
>
> Last column mem_used is to indicate how many allocator used the memory
> to store compressed page
>
> 1) compression ratio 75
>
> WRITE 2535 WRITE 1928
> WRITE 2425 WRITE 1886
> READ 6211 READ 5731
> READ 6339 READ 6182
> READ 1791 READ 1592
> WRITE 1790 WRITE 1591
> READ 1704 READ 1493
> WRITE 1699 WRITE 1489
> WRITE 984 WRITE 974
> TRIM 984 TRIM 974
> mem_used 29986816 mem_used 61239296
>
> For every operation, zsmalloc is faster than z3fold.
> Even, it used the 1/2 memory compared to z3fold.
>
> 2) compression ratio 66
>
> WRITE 2125 WRITE 1258
> WRITE 2107 WRITE 1233
> READ 5714 READ 5793
> READ 5948 READ 6065
> READ 1667 READ 1248
> WRITE 1666 WRITE 1247
> READ 1521 READ 1218
> WRITE 1517 WRITE 1215
> WRITE 943 WRITE 870
> TRIM 943 TRIM 870
> mem_used 38158336 mem_used 76779520
>
> For only read operation, z3fold is a bit faster than zsmalloc about 2%.
> However, look at other operations which zsmalloc is much faster.
> Even, look at used memory.
>
> 3) compression ratio 50
>
> WRITE 2051 WRITE 1109
> WRITE 2029 WRITE 1087
> READ 5366 READ 6364
> READ 5575 READ 5785
> READ 1497 READ 1121
> WRITE 1496 WRITE 1121
> READ 1432 READ 1065
> WRITE 1428 WRITE 1062
> WRITE 930 WRITE 838
> TRIM 930 TRIM 838
> mem_used 59932672 mem_used 104873984
>
> sequential read on z3fold is faster about 15%. However, look at other
> operations and used memory. zsmalloc is better.
There are two things to this: the measurements you've taken as such
and how they are relevant to this discussion.
I'd be happy to discuss these measurements in a separate thread if you
specified more precisely what kind of x86 the measurements were taken
on.
However, my point was that there are rather common cases when people
want to use z3fold as a zRAM memory allocation backend. The fact that
there are other cases when people wouldn't want that is pretty natural
and doesn't need a proof.
That's why I propose to use ZRAM over zpool API for the sake of
flexibility. That would benefit various users of ZRAM and, at the end
of the day, the Linux kernel ecosystem.
<snip>
> Thanks for the testing. I also tried to test zbud with zram but failed because fio
> submit incompressible pages to zram even though it specifiy compress ratio 100%
> However, zbud doesn't support 4K page allocation so zram couldn't work on it
> at this moment. I tried various fio versions as well as old but everything failed.
>
> How did you test it successfully? Let me know your fio version.
> I want to investigate what's the performance bottleneck beside page copy
> so that I will optimize it.
You're very welcome. :) The patch to make zbud accept PAGE_SIZE pages
has been posted a while ago [1] and it was a part of our previous
(pre-z3fold) discussion on the same subject but you probably haven't
read it then.
> >
> > Now to the fun part.
> > zsmalloc:
> > 0 .text 00002908 0000000000000000 0000000000000000 00000040 2**2
> > CONTENTS, ALLOC, LOAD, RELOC, READONLY, CODE
> > zbud:
> > 0 .text 0000072c 0000000000000000 0000000000000000 00000040 2**2
> > CONTENTS, ALLOC, LOAD, RELOC, READONLY, CODE
> >
> > And this does not cover dynamic memory allocation overhead which is
> > higher for zsmalloc. So once again, given that the compression ratio
> > is low (e. g. a simple HW accelerator is used), what would most
> > unbiased people prefer to use in this case?
>
> Zsmalloc has more features than zbud. That's why you see the code size
> difference. It was intentional because at that time most of users were
> mobile phones, TV and other smart devices. They needed those features.
>
> We could make those feature turned off at build time, which will improve
> performance and reduce code size a lot. It would be no problem if the
> user wanted to use zbud which is alredy lacking of those features.
I do support this idea and would like to help as much as I can, but
why should the people who want to use ZRAM/zbud combo be left stranded
while we're working on reducing the zsmalloc code size by 4x?
With that said, let me also re-iterate that there may be more
allocators coming, and in some cases zsmalloc won't be a good
fit/alternative while there will be still a need for a compressed RAM
device. I hope you understand.
Best regards,
Vitaly
[1] https://lore.kernel.org/patchwork/patch/598210/