Re: [PATCH] squashfs: enhance parallel I/O

[Minchan Kim]

Hello Phillip,

Sorry for late response. I'm still in Edinburgh.

On Wed, Oct 23, 2013 at 07:21:39AM +0100, Phillip Lougher wrote:
> 
> Hi Minchan,
> 
> Apologies for the lateness of this review, I had forgotten I'd not
> send it.
> 
> Minchan Kim <minchan@xxxxxxxxxx> wrote:
> >Now squashfs have used for only one stream buffer for decompression
> >so it hurts concurrent read performance so this patch supprts
> >multiple decompressor to enhance performance concurrent I/O.
> >
> >Four 1G file dd read on KVM machine which has 2 CPU and 4G memory.
> >
> >dd if=test/test1.dat of=/dev/null &
> >dd if=test/test2.dat of=/dev/null &
> >dd if=test/test3.dat of=/dev/null &
> >dd if=test/test4.dat of=/dev/null &
> >
> >old : 1m39s -> new : 9s
> >
> >This patch is based on [1].
> >
> >[1] Squashfs: Refactor decompressor interface and code
> >
> >Cc: Phillip Lougher <phillip@xxxxxxxxxxxxxxx>
> >Signed-off-by: Minchan Kim <minchan@xxxxxxxxxx>
> >
> >---
> >fs/squashfs/Kconfig              |   12 +++
> >fs/squashfs/Makefile             |    9 +-
> >fs/squashfs/decompressor_multi.c |  194 ++++++++++++++++++++++++++++++++++++++
> >3 files changed, 214 insertions(+), 1 deletion(-)
> >create mode 100644 fs/squashfs/decompressor_multi.c
> >
> >diff --git a/fs/squashfs/Kconfig b/fs/squashfs/Kconfig
> >index c70111e..7a580d0 100644
> >--- a/fs/squashfs/Kconfig
> >+++ b/fs/squashfs/Kconfig
> >@@ -63,6 +63,18 @@ config SQUASHFS_LZO
> >
> >	  If unsure, say N.
> >
> >+config SQUASHFS_MULTI_DECOMPRESSOR
> 
> Small alterations to the English, I don't like doing this because
> English is probably a foreign language to you, and my Korean is non-existent!
> but, this is user visible and you did say you wanted review !

Yeah, I'm really welcome to fix my English and it's really power of open source
community where native and non-native people could help each other.

> 
> >+	bool "Use multiple decompressor for handling concurrent I/O"
> 
> bool "Use multiple decompressors for handling parallel I/O"
> 
> Concurrent is subtly different to parallel, and you use parallel in
> the following description.
> 
> 
> >+	depends on SQUASHFS
> >+	help
> >+	  By default Squashfs uses a compressor for data but it gives
> >+	  poor performance on parallel I/O workload of multiple CPU
> >+	  mahchine due to waitting a compressor available.
> >+
> 
> By default Squashfs uses a single decompressor but it gives
> poor performance on parallel I/O workloads when using multiple CPU
> machines due to waiting on decompressor availability.
> 
> >+	  If workload has parallel I/O and your system has enough memory,
> >+	  this option may improve overall I/O performance.
> >+	  If unsure, say N.
> >+
> 
> If you have a parallel I/O workload and your system has enough memory,
> using this option may improve overall I/O performance.
> 
> If unsure, say N.

Will correct.

> >+
> >config SQUASHFS_XZ
> >	bool "Include support for XZ compressed file systems"
> >	depends on SQUASHFS
> >diff --git a/fs/squashfs/Makefile b/fs/squashfs/Makefile
> >index c223c84..dfebc3b 100644
> >--- a/fs/squashfs/Makefile
> >+++ b/fs/squashfs/Makefile
> >@@ -4,8 +4,15 @@
> >
> >obj-$(CONFIG_SQUASHFS) += squashfs.o
> >squashfs-y += block.o cache.o dir.o export.o file.o fragment.o id.o inode.o
> >-squashfs-y += namei.o super.o symlink.o decompressor.o decompressor_single.o
> >+squashfs-y += namei.o super.o symlink.o decompressor.o
> >+
> >squashfs-$(CONFIG_SQUASHFS_XATTR) += xattr.o xattr_id.o
> >squashfs-$(CONFIG_SQUASHFS_LZO) += lzo_wrapper.o
> >squashfs-$(CONFIG_SQUASHFS_XZ) += xz_wrapper.o
> >squashfs-$(CONFIG_SQUASHFS_ZLIB) += zlib_wrapper.o
> >+
> >+ifdef CONFIG_SQUASHFS_MULTI_DECOMPRESSOR
> >+	squashfs-y		+= decompressor_multi.o
> >+else
> >+	squashfs-y		+= decompressor_single.o
> >+endif
> >diff --git a/fs/squashfs/decompressor_multi.c b/fs/squashfs/decompressor_multi.c
> >new file mode 100644
> >index 0000000..23f1e94
> >--- /dev/null
> >+++ b/fs/squashfs/decompressor_multi.c
> >@@ -0,0 +1,194 @@
> >+/*
> >+ *  Copyright (c) 2013
> >+ *  Minchan Kim <minchan@xxxxxxxxxx>
> >+ *
> >+ *  This work is licensed under the terms of the GNU GPL, version 2. See
> >+ *  the COPYING file in the top-level directory.
> >+ */
> >+#include <linux/types.h>
> >+#include <linux/mutex.h>
> >+#include <linux/slab.h>
> >+#include <linux/buffer_head.h>
> >+#include <linux/sched.h>
> >+#include <linux/wait.h>
> >+#include <linux/cpumask.h>
> >+
> >+#include "squashfs_fs.h"
> >+#include "squashfs_fs_sb.h"
> >+#include "decompressor.h"
> >+#include "squashfs.h"
> >+
> >+/*
> >+ * This file implements multi-threaded decompression in the
> >+ * decompressor framework
> >+ */
> >+
> >+
> >+/*
> >+ * The reason that multiply two is that a CPU can request new I/O
> >+ * while it is waitting previous request.
> 
> s/waitting/waiting/

Oops.

> 
> The English in the comments is understandable, and not user-visible, and
> so I won't change it, except for spelling mistakes ...

I understand you respect my work although it couldn't satisfy your high bar
so I don't care if you correct this. :)

> 
> >+ */
> >+#define MAX_DECOMPRESSOR	(num_online_cpus() * 2)
> >+
> >+
> >+int squashfs_max_decompressors(void)
> >+{
> >+	return MAX_DECOMPRESSOR;
> >+}
> >+
> >+
> >+struct squashfs_stream {
> >+	void			*comp_opts;
> >+	struct list_head	strm_list;
> >+	struct mutex		mutex;
> >+	int			avail_comp;
> >+	wait_queue_head_t	wait;
> >+};
> >+
> >+
> >+struct comp_stream {
> >+	void *stream;
> >+	struct list_head list;
> >+};
> >+
> 
> One general small nitpick, you use "comp" to name things, comp_stream,
> avail_comp etc.  But, as this is a decompressor, it should more correctly
> be decomp...

Will change.

> 
> (note comp_opts is not decomp_opts because it is the compression options
> selected by the user at compression time) ...

True.

> 
> >+
> >+static void put_comp_stream(struct comp_stream *comp_strm,
> >+				struct squashfs_stream *stream)
> >+{
> >+	mutex_lock(&stream->mutex);
> >+	list_add(&comp_strm->list, &stream->strm_list);
> >+	mutex_unlock(&stream->mutex);
> >+	wake_up(&stream->wait);
> >+}
> >+
> >+void *squashfs_decompressor_create(struct squashfs_sb_info *msblk,
> >+				void *comp_opts)
> >+{
> >+	struct squashfs_stream *stream;
> >+	struct comp_stream *comp_strm = NULL;
> >+	int err = -ENOMEM;
> >+
> >+	stream = kzalloc(sizeof(*stream), GFP_KERNEL);
> >+	if (!stream)
> >+		goto out;
> >+
> >+	stream->comp_opts = comp_opts;
> >+	mutex_init(&stream->mutex);
> >+	INIT_LIST_HEAD(&stream->strm_list);
> >+	init_waitqueue_head(&stream->wait);
> >+
> >+	comp_strm = kmalloc(sizeof(*comp_strm), GFP_KERNEL);
> >+	if (!comp_strm)
> >+		goto out;
> >+
> >+	comp_strm->stream = msblk->decompressor->init(msblk,
> >+						stream->comp_opts);
> >+	if (IS_ERR(comp_strm->stream)) {
> >+		err = PTR_ERR(comp_strm->stream);
> >+		goto out;
> >+	}
> >+
> >+	list_add(&comp_strm->list, &stream->strm_list);
> >+	stream->avail_comp = 1;
> 
> I assume you're always creating a decompressor here (rather than
> setting it to 0, and allocating the first one in get_comp_stream) to
> ensure there's always at least one decompressor available going into
> get_comp_stream()?  So if decompressor creation fails in
> get_comp_stream() we can always fall back to using the decompressors
> already allocated, because we know there will always be at least one?

Right.

> 
> Maybe a comment to that effect? To show creating a decompressor here
> and setting this to one is deliberate....  This will prevent others
> from thinking they can "optimise" the code by having the first decompressor
> created in get_comp_stream()!

Indeed. I will add a comment about that. Of course you could feel free to
fix English if it doesn't meet your bar and I will learn English from native
people without any charge. I's really nice thing for me.

> 
> /* ensure we have at least one decompressor in get_comp_stream() */ ?
> 
> >+	return stream;
> >+
> >+out:
> >+	kfree(comp_strm);
> >+	kfree(stream);
> >+	return ERR_PTR(err);
> >+}
> >+
> >+
> >+void squashfs_decompressor_destroy(struct squashfs_sb_info *msblk)
> >+{
> >+	struct squashfs_stream *stream = msblk->stream;
> >+	if (stream) {
> >+		struct comp_stream *comp_strm;
> >+
> >+		while (!list_empty(&stream->strm_list)) {
> >+			comp_strm = list_entry(stream->strm_list.prev,
> >+						struct comp_stream, list);
> >+			list_del(&comp_strm->list);
> >+			msblk->decompressor->free(comp_strm->stream);
> >+			kfree(comp_strm);
> >+			stream->avail_comp--;
> >+		}
> >+	}
> >+
> >+	WARN_ON(stream->avail_comp);
> >+	kfree(stream->comp_opts);
> >+	kfree(stream);
> >+}
> >+
> >+
> >+static struct comp_stream *get_comp_stream(struct squashfs_sb_info *msblk,
> >+					struct squashfs_stream *stream)
> >+{
> >+	struct comp_stream *comp_strm;
> >+
> >+	while (1) {
> >+		mutex_lock(&stream->mutex);
> >+
> >+		/* There is available comp_stream */
> >+		if (!list_empty(&stream->strm_list)) {
> >+			comp_strm = list_entry(stream->strm_list.prev,
> >+				struct comp_stream, list);
> >+			list_del(&comp_strm->list);
> >+			mutex_unlock(&stream->mutex);
> >+			break;
> >+		}
> >+
> >+		/*
> >+		 * If there is no available comp and already full,
> >+		 * let's wait for releasing comp from other users.
> >+		 */
> >+		if (stream->avail_comp >= MAX_DECOMPRESSOR)
> >+			goto wait;
> >+
> >+		/* Let's allocate new comp */
> >+		comp_strm = kmalloc(sizeof(*comp_strm), GFP_KERNEL);
> >+		if (!comp_strm)
> >+			goto wait;
> >+
> >+		comp_strm->stream = msblk->decompressor->init(msblk,
> >+						stream->comp_opts);
> >+		if (IS_ERR(comp_strm->stream)) {
> >+			kfree(comp_strm);
> >+			goto wait;
> >+		}
> >+
> >+		stream->avail_comp++;
> >+		WARN_ON(stream->avail_comp > MAX_DECOMPRESSOR);
> >+
> >+		mutex_unlock(&stream->mutex);
> >+		break;
> >+wait:
> >+		/*
> >+		 * If system memory is tough, let's for other's
> >+		 * releasing instead of hurting VM because it could
> >+		 * make page cache thrashing.
> >+		 */
> 
> 
> >+		mutex_unlock(&stream->mutex);
> >+		wait_event(stream->wait,
> >+			!list_empty(&stream->strm_list));
> >+	}
> >+
> >+	return comp_strm;
> >+}
> >+
> >+
> >+int squashfs_decompress(struct squashfs_sb_info *msblk,
> >+	void **buffer, struct buffer_head **bh, int b, int offset, int length,
> >+	int srclength, int pages)
> >+{
> 
> The interface here is changed by the introduction of the page
> handler abstraction...
> 
> I can apply your patch after the refactoring patch and before the
> "Squashfs: Directly decompress into the page cache for file" patch-set
> and fix up the code here....  Or you can fix up the code?  I'm
> happy to do either, whatever you prefer.

I think this stuff is more simpler than "directly decompress" patch
so I hope let's merge this first before "directly deompress" because
it would make git-bisect/revert more simple if "directly decompress"
patch might make a problem.

I will send fixup patch as soon as I return to office in Korea.

Thanks for the your help!


> 
> Thanks
> 
> Phillip
> 
> 
> >+	int res;
> >+	struct squashfs_stream *stream = msblk->stream;
> >+	struct comp_stream *comp_stream = get_comp_stream(msblk, stream);
> >+	res = msblk->decompressor->decompress(msblk, comp_stream->stream,
> >+		buffer, bh, b, offset, length, srclength, pages);
> >+	put_comp_stream(comp_stream, stream);
> >+	if (res < 0)
> >+		ERROR("%s decompression failed, data probably corrupt\n",
> >+			msblk->decompressor->name);
> >+	return res;
> >+}

-- 
Kind regards,
Minchan Kim
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/