Skip to contentSkip to conversation

Use BIO_META tags for metadata rather than BIO_SYNC

PATCH In review David Chinner 3 participants 3 messages Jan 21, 2008 – Jul 23, 2008

Subsystems (2)
VFS XFS filesystem
lore
Share & tools
lore

David Chinner

author
#0 lore raw

PATCH 1/1 Use BIO_META tags for metadata rather than BIO_SYNC

Improve metadata I/O merging in the elevator

Change all async metadata buffers to use [READ|WRITE]_META I/O types so that the I/O doesn't get issued immediately. This allows merging of adjacent metadata requests but still prioritises them over bulk data. This shows a 10-15% improvement in sequential create speed of small files.

Don't include the log buffers in this classification - leave them as sync types so they are issued immediately.

Signed-off-by Dave Chinner <[email protected]>
---
 fs/xfs/linux-2.6/xfs_buf.c |    6 +++++-
 include/linux/fs.h         |    1 +
 2 files changed, 6 insertions(+), 1 deletion(-)
fs/xfs/linux-2.6/xfs_buf.c +5 -1
Index: 2.6.x-xfs-new/fs/xfs/linux-2.6/xfs_buf.c
===================================================================
@@ -1136,10 +1136,14 @@ _xfs_buf_ioapply(
 	if (bp->b_flags & XBF_ORDERED) {
 		ASSERT(!(bp->b_flags & XBF_READ));
 		rw = WRITE_BARRIER;
-	} else if (bp->b_flags & _XBF_RUN_QUEUES) {
+	} else if (bp->b_flags & XBF_LOG_BUFFER) {
 		ASSERT(!(bp->b_flags & XBF_READ_AHEAD));
 		bp->b_flags &= ~_XBF_RUN_QUEUES;
 		rw = (bp->b_flags & XBF_WRITE) ? WRITE_SYNC : READ_SYNC;
+	} else if (bp->b_flags & _XBF_RUN_QUEUES) {
+		ASSERT(!(bp->b_flags & XBF_READ_AHEAD));
+		bp->b_flags &= ~_XBF_RUN_QUEUES;
+		rw = (bp->b_flags & XBF_WRITE) ? WRITE_META : READ_META;
 	} else {
 		rw = (bp->b_flags & XBF_WRITE) ? WRITE :
 		     (bp->b_flags & XBF_READ_AHEAD) ? READA : READ;
include/linux/fs.h +1 -0
Index: 2.6.x-xfs-new/include/linux/fs.h
===================================================================
@@ -83,6 +83,7 @@ extern int dir_notify_enable;
 #define READ_SYNC	(READ | (1 << BIO_RW_SYNC))
 #define READ_META	(READ | (1 << BIO_RW_META))
 #define WRITE_SYNC	(WRITE | (1 << BIO_RW_SYNC))
+#define WRITE_META	(WRITE | (1 << BIO_RW_META))
 #define WRITE_BARRIER	((1 << BIO_RW) | (1 << BIO_RW_BARRIER))
 
 #define SEL_IN		1
#1 lore raw

Looks good.

Quoted 52 lines · David Chinner

Improve metadata I/O merging in the elevator

Change all async metadata buffers to use [READ|WRITE]_META I/O types so that the I/O doesn't get issued immediately. This allows merging of adjacent metadata requests but still prioritises them over bulk data. This shows a 10-15% improvement in sequential create speed of small files.

Don't include the log buffers in this classification - leave them as sync types so they are issued immediately.

Signed-off-by Dave Chinner <[email protected]>
---
 fs/xfs/linux-2.6/xfs_buf.c |    6 +++++-
 include/linux/fs.h         |    1 +
 2 files changed, 6 insertions(+), 1 deletion(-)

Index: 2.6.x-xfs-new/fs/xfs/linux-2.6/xfs_buf.c
===================================================================

fs/xfs/linux-2.6/xfs_buf.c +5 -1
@@ -1136,10 +1136,14 @@ _xfs_buf_ioapply(
 	if (bp->b_flags & XBF_ORDERED) {
 		ASSERT(!(bp->b_flags & XBF_READ));
 		rw = WRITE_BARRIER;
-	} else if (bp->b_flags & _XBF_RUN_QUEUES) {
+	} else if (bp->b_flags & XBF_LOG_BUFFER) {
 		ASSERT(!(bp->b_flags & XBF_READ_AHEAD));
 		bp->b_flags &= ~_XBF_RUN_QUEUES;
 		rw = (bp->b_flags & XBF_WRITE) ? WRITE_SYNC : READ_SYNC;
+	} else if (bp->b_flags & _XBF_RUN_QUEUES) {
+		ASSERT(!(bp->b_flags & XBF_READ_AHEAD));
+		bp->b_flags &= ~_XBF_RUN_QUEUES;
+		rw = (bp->b_flags & XBF_WRITE) ? WRITE_META : READ_META;
 	} else {
 		rw = (bp->b_flags & XBF_WRITE) ? WRITE :
 		     (bp->b_flags & XBF_READ_AHEAD) ? READA : READ;

Index: 2.6.x-xfs-new/include/linux/fs.h
===================================================================

include/linux/fs.h +1 -0
@@ -83,6 +83,7 @@ extern int dir_notify_enable;
 #define READ_SYNC	(READ | (1 << BIO_RW_SYNC))
 #define READ_META	(READ | (1 << BIO_RW_META))
 #define WRITE_SYNC	(WRITE | (1 << BIO_RW_SYNC))
+#define WRITE_META	(WRITE | (1 << BIO_RW_META))
 #define WRITE_BARRIER	((1 << BIO_RW) | (1 << BIO_RW_BARRIER))
 
 #define SEL_IN		1
#2 lore raw
In reply to David Chinner
Quoted 7 lines · David Chinner

Improve metadata I/O merging in the elevator

Change all async metadata buffers to use [READ|WRITE]_META I/O types so that the I/O doesn't get issued immediately. This allows merging of adjacent metadata requests but still prioritises them over bulk data. This shows a 10-15% improvement in sequential create speed of small files.

That sounds awesome. I'll test it on a similar workload I've been playing with. Lachlan reviewed this, I think - any plans to add it?

I'll run my test & post results.

Thanks,
-Eric

Quoted 43 lines

Don't include the log buffers in this classification - leave them as sync types so they are issued immediately.

Signed-off-by Dave Chinner <[email protected]>
---
 fs/xfs/linux-2.6/xfs_buf.c |    6 +++++-
 include/linux/fs.h         |    1 +
 2 files changed, 6 insertions(+), 1 deletion(-)

Index: 2.6.x-xfs-new/fs/xfs/linux-2.6/xfs_buf.c
===================================================================

fs/xfs/linux-2.6/xfs_buf.c +5 -1
@@ -1136,10 +1136,14 @@ _xfs_buf_ioapply(
 	if (bp->b_flags & XBF_ORDERED) {
 		ASSERT(!(bp->b_flags & XBF_READ));
 		rw = WRITE_BARRIER;
-	} else if (bp->b_flags & _XBF_RUN_QUEUES) {
+	} else if (bp->b_flags & XBF_LOG_BUFFER) {
 		ASSERT(!(bp->b_flags & XBF_READ_AHEAD));
 		bp->b_flags &= ~_XBF_RUN_QUEUES;
 		rw = (bp->b_flags & XBF_WRITE) ? WRITE_SYNC : READ_SYNC;
+	} else if (bp->b_flags & _XBF_RUN_QUEUES) {
+		ASSERT(!(bp->b_flags & XBF_READ_AHEAD));
+		bp->b_flags &= ~_XBF_RUN_QUEUES;
+		rw = (bp->b_flags & XBF_WRITE) ? WRITE_META : READ_META;
 	} else {
 		rw = (bp->b_flags & XBF_WRITE) ? WRITE :
 		     (bp->b_flags & XBF_READ_AHEAD) ? READA : READ;

Index: 2.6.x-xfs-new/include/linux/fs.h
===================================================================

include/linux/fs.h +1 -0
@@ -83,6 +83,7 @@ extern int dir_notify_enable;
 #define READ_SYNC	(READ | (1 << BIO_RW_SYNC))
 #define READ_META	(READ | (1 << BIO_RW_META))
 #define WRITE_SYNC	(WRITE | (1 << BIO_RW_SYNC))
+#define WRITE_META	(WRITE | (1 << BIO_RW_META))
 #define WRITE_BARRIER	((1 << BIO_RW) | (1 << BIO_RW_BARRIER))
 
 #define SEL_IN		1