branches/zip: Replace the constant 3/8 ratio that controls the LRU_old

size with the settable global variable innodb_old_blocks_pct. The
minimum and maximum values are 5 and 95 per cent, respectively. The
default is 100*3/8, in line with the old behavior.

ut_time_ms(): New utility function, to return the current time in
milliseconds. TODO: Is there a more efficient timestamp function, such
as rdtsc divided by a power of two?

buf_LRU_old_threshold_ms: New variable, corresponding to
innodb_old_blocks_time. The value 0 is the default behaviour: no
timeout before making blocks 'new'.

bpage->accessed, bpage->LRU_position, buf_pool->ulint_clock: Remove.

bpage->access_time: New field, replacing bpage->accessed. Protected by
buf_pool_mutex instead of bpage->mutex. Updated when a page is created
or accessed the first time in the buffer pool.

buf_LRU_old_ratio, innobase_old_blocks_pct: New variables,
corresponding to innodb_old_blocks_pct

buf_LRU_old_ratio_update(), innobase_old_blocks_pct_update(): Update
functions for buf_LRU_old_ratio, innobase_old_blocks_pct.

buf_page_peek_if_too_old(): Compare ut_time_ms() to bpage->access_time
if buf_LRU_old_threshold_ms && bpage->old.  Else observe
buf_LRU_old_ratio and bpage->freed_page_clock.

buf_pool_t: Add n_pages_made_young, n_pages_not_made_young,
n_pages_made_young_old, n_pages_not_made_young, for statistics.

buf_print(): Display buf_pool->n_pages_made_young,
buf_pool->n_pages_not_made_young.  This function is only for crash
diagnostics.

buf_print_io(): Display buf_pool->LRU_old_len and quantities derived
from buf_pool->n_pages_made_young, buf_pool->n_pages_not_made_young.
This function is invoked by SHOW ENGINE INNODB STATUS.

rb://129 approved by Heikki Tuuri.  This addresses Bug #45015.

ChangeLog

+2009-08-27	The InnoDB Team
+
+	* buf/buf0buf.c, buf/buf0lru.c, buf/buf0rea.c,
+	handler/ha_innodb.cc, include/buf0buf.h, include/buf0buf.ic,
+	include/buf0lru.h, include/ut0ut.h, ut/ut0ut.c:
+	Make it possible to tune the buffer pool LRU eviction policy to be
+	more resistant against index scans.  Introduce the settable global
+	variables innodb_old_blocks_pct and innodb_old_blocks_time for
+	controlling the buffer pool eviction policy.  The parameter
+	innodb_old_blocks_pct (5..95) controls the desired amount of "old"
+	blocks in the LRU list.  The default is 37, corresponding to the
+	old fixed ratio of 3/8.  Each time a block is accessed, it will be
+	moved to the "new" blocks if its first access was at least
+	innodb_old_blocks_time milliseconds ago (default 0, meaning every
+	block).  The idea is that in index scans, blocks will be accessed
+	a few times within innodb_old_blocks_time, and they will remain in
+	the "old" section of the LRU list.  Thus, when
+	innodb_old_blocks_time is nonzero, blocks retrieved for one-time
+	index scans will be more likely candidates for eviction than
+	blocks that are accessed in random patterns.
+
 2009-08-26	The InnoDB Team
 
 	* handler/ha_innodb.cc, os/os0file.c:

buf/buf0buf.c

@@ -966,8 +966,6 @@ buf_pool_init(void)
 		buf_pool->no_flush[i] = os_event_create(NULL);
 	}
 
-	buf_pool->ulint_clock = 1;
-
 	/* 3. Initialize LRU fields
 	--------------------------- */
 	/* All fields are initialized by mem_zalloc(). */
@@ -1470,31 +1468,6 @@ buf_pool_resize(void)
 	buf_pool_page_hash_rebuild();
 }
 
-/********************************************************************//**
-Moves the block to the start of the LRU list if there is a danger
-that the block would drift out of the buffer pool. */
-UNIV_INLINE
-void
-buf_block_make_young(
-/*=================*/
-	buf_page_t*	bpage)	/*!< in: block to make younger */
-{
-	ut_ad(!buf_pool_mutex_own());
-
-	/* Note that we read freed_page_clock's without holding any mutex:
-	this is allowed since the result is used only in heuristics */
-
-	if (buf_page_peek_if_too_old(bpage)) {
-
-		buf_pool_mutex_enter();
-		/* There has been freeing activity in the LRU list:
-		best to move to the head of the LRU list */
-
-		buf_LRU_make_block_young(bpage);
-		buf_pool_mutex_exit();
-	}
-}
-
 /********************************************************************//**
 Moves a page to the start of the buffer pool LRU list. This high-level
 function can be used to prevent an important page from from slipping out of
@@ -1713,13 +1686,15 @@ err_exit:
 got_block:
 	must_read = buf_page_get_io_fix(bpage) == BUF_IO_READ;
 
-	buf_pool_mutex_exit();
+	if (buf_page_peek_if_too_old(bpage)) {
+		buf_LRU_make_block_young(bpage);
+	}
 
-	buf_page_set_accessed(bpage, TRUE);
+	buf_page_set_accessed(bpage);
 
-	mutex_exit(block_mutex);
+	buf_pool_mutex_exit();
 
-	buf_block_make_young(bpage);
+	mutex_exit(block_mutex);
 
 #ifdef UNIV_DEBUG_FILE_ACCESSES
 	ut_a(!bpage->file_page_was_freed);
@@ -2000,7 +1975,7 @@ buf_page_get_gen(
 	mtr_t*		mtr)	/*!< in: mini-transaction */
 {
 	buf_block_t*	block;
-	ibool		accessed;
+	unsigned	access_time;
 	ulint		fix_type;
 	ibool		must_read;
 
@@ -2243,17 +2218,20 @@ wait_until_unfixed:
 	UNIV_MEM_ASSERT_RW(&block->page, sizeof block->page);
 
 	buf_block_buf_fix_inc(block, file, line);
-	buf_pool_mutex_exit();
+
+	mutex_exit(&block->mutex);
 
 	/* Check if this is the first access to the page */
 
-	accessed = buf_page_is_accessed(&block->page);
+	access_time = buf_page_is_accessed(&block->page);
 
-	buf_page_set_accessed(&block->page, TRUE);
+	if (buf_page_peek_if_too_old(&block->page)) {
+		buf_LRU_make_block_young(&block->page);
+	}
 
-	mutex_exit(&block->mutex);
+	buf_page_set_accessed(&block->page);
 
-	buf_block_make_young(&block->page);
+	buf_pool_mutex_exit();
 
 #ifdef UNIV_DEBUG_FILE_ACCESSES
 	ut_a(!block->page.file_page_was_freed);
@@ -2306,7 +2284,7 @@ wait_until_unfixed:
 
 	mtr_memo_push(mtr, block, fix_type);
 
-	if (!accessed) {
+	if (!access_time) {
 		/* In the case of a first access, try to apply linear
 		read-ahead */
 
@@ -2336,7 +2314,7 @@ buf_page_optimistic_get_func(
 	ulint		line,	/*!< in: line where called */
 	mtr_t*		mtr)	/*!< in: mini-transaction */
 {
-	ibool		accessed;
+	unsigned	access_time;
 	ibool		success;
 	ulint		fix_type;
 
@@ -2353,14 +2331,21 @@ buf_page_optimistic_get_func(
 	}
 
 	buf_block_buf_fix_inc(block, file, line);
-	accessed = buf_page_is_accessed(&block->page);
-	buf_page_set_accessed(&block->page, TRUE);
 
 	mutex_exit(&block->mutex);
 
-	buf_block_make_young(&block->page);
+	buf_pool_mutex_enter();
 
 	/* Check if this is the first access to the page */
+	access_time = buf_page_is_accessed(&block->page);
+
+	if (buf_page_peek_if_too_old(&block->page)) {
+		buf_LRU_make_block_young(&block->page);
+	}
+
+	buf_page_set_accessed(&block->page);
+
+	buf_pool_mutex_exit();
 
 	ut_ad(!ibuf_inside()
 	      || ibuf_page(buf_block_get_space(block),
@@ -2412,7 +2397,7 @@ buf_page_optimistic_get_func(
 #ifdef UNIV_DEBUG_FILE_ACCESSES
 	ut_a(block->page.file_page_was_freed == FALSE);
 #endif
-	if (UNIV_UNLIKELY(!accessed)) {
+	if (UNIV_UNLIKELY(!access_time)) {
 		/* In the case of a first access, try to apply linear
 		read-ahead */
 
@@ -2473,10 +2458,16 @@ buf_page_get_known_nowait(
 
 	mutex_exit(&block->mutex);
 
-	if (mode == BUF_MAKE_YOUNG) {
-		buf_block_make_young(&block->page);
+	buf_pool_mutex_enter();
+
+	if (mode == BUF_MAKE_YOUNG && buf_page_peek_if_too_old(&block->page)) {
+		buf_LRU_make_block_young(&block->page);
 	}
 
+	buf_page_set_accessed(&block->page);
+
+	buf_pool_mutex_exit();
+
 	ut_ad(!ibuf_inside() || (mode == BUF_KEEP_OLD));
 
 	if (rw_latch == RW_S_LATCH) {
@@ -2608,10 +2599,10 @@ buf_page_init_low(
 	buf_page_t*	bpage)	/*!< in: block to init */
 {
 	bpage->flush_type = BUF_FLUSH_LRU;
-	bpage->accessed = FALSE;
 	bpage->io_fix = BUF_IO_NONE;
 	bpage->buf_fix_count = 0;
 	bpage->freed_page_clock = 0;
+	bpage->access_time = 0;
 	bpage->newest_modification = 0;
 	bpage->oldest_modification = 0;
 	HASH_INVALIDATE(bpage, hash);
@@ -2990,12 +2981,12 @@ buf_page_create(
 		rw_lock_x_unlock(&block->lock);
 	}
 
+	buf_page_set_accessed(&block->page);
+
 	buf_pool_mutex_exit();
 
 	mtr_memo_push(mtr, block, MTR_MEMO_BUF_FIX);
 
-	buf_page_set_accessed(&block->page, TRUE);
-
 	mutex_exit(&block->mutex);
 
 	/* Delete possible entries for the page from the insert buffer:
@@ -3528,6 +3519,7 @@ buf_print(void)
 		"n pending decompressions %lu\n"
 		"n pending reads %lu\n"
 		"n pending flush LRU %lu list %lu single page %lu\n"
+		"pages made young %lu, not young %lu\n"
 		"pages read %lu, created %lu, written %lu\n",
 		(ulong) size,
 		(ulong) UT_LIST_GET_LEN(buf_pool->LRU),
@@ -3538,6 +3530,8 @@ buf_print(void)
 		(ulong) buf_pool->n_flush[BUF_FLUSH_LRU],
 		(ulong) buf_pool->n_flush[BUF_FLUSH_LIST],
 		(ulong) buf_pool->n_flush[BUF_FLUSH_SINGLE_PAGE],
+		(ulong) buf_pool->n_pages_made_young,
+		(ulong) buf_pool->n_pages_not_made_young,
 		(ulong) buf_pool->n_pages_read, buf_pool->n_pages_created,
 		(ulong) buf_pool->n_pages_written);
 
@@ -3744,10 +3738,9 @@ buf_print_io(
 {
 	time_t	current_time;
 	double	time_elapsed;
-	ulint	size;
+	ulint	n_gets_diff;
 
 	ut_ad(buf_pool);
-	size = buf_pool->curr_size;
 
 	buf_pool_mutex_enter();
 
@@ -3755,12 +3748,14 @@ buf_print_io(
 		"Buffer pool size   %lu\n"
 		"Free buffers       %lu\n"
 		"Database pages     %lu\n"
+		"Old database pages %lu\n"
 		"Modified db pages  %lu\n"
 		"Pending reads %lu\n"
 		"Pending writes: LRU %lu, flush list %lu, single page %lu\n",
-		(ulong) size,
+		(ulong) buf_pool->curr_size,
 		(ulong) UT_LIST_GET_LEN(buf_pool->free),
 		(ulong) UT_LIST_GET_LEN(buf_pool->LRU),
+		(ulong) buf_pool->LRU_old_len,
 		(ulong) UT_LIST_GET_LEN(buf_pool->flush_list),
 		(ulong) buf_pool->n_pend_reads,
 		(ulong) buf_pool->n_flush[BUF_FLUSH_LRU]
@@ -3775,8 +3770,18 @@ buf_print_io(
 	buf_pool->last_printout_time = current_time;
 
 	fprintf(file,
+		"Pages made young %lu, not young %lu\n"
+		"%.2f youngs/s, %.2f non-youngs/s\n"
 		"Pages read %lu, created %lu, written %lu\n"
 		"%.2f reads/s, %.2f creates/s, %.2f writes/s\n",
+		(ulong) buf_pool->n_pages_made_young,
+		(ulong) buf_pool->n_pages_not_made_young,
+		(buf_pool->n_pages_made_young
+		 - buf_pool->n_pages_made_young_old)
+		/ time_elapsed,
+		(buf_pool->n_pages_not_made_young
+		 - buf_pool->n_pages_not_made_young_old)
+		/ time_elapsed,
 		(ulong) buf_pool->n_pages_read,
 		(ulong) buf_pool->n_pages_created,
 		(ulong) buf_pool->n_pages_written,
@@ -3787,19 +3792,33 @@ buf_print_io(
 		(buf_pool->n_pages_written - buf_pool->n_pages_written_old)
 		/ time_elapsed);
 
-	if (buf_pool->n_page_gets > buf_pool->n_page_gets_old) {
-		fprintf(file, "Buffer pool hit rate %lu / 1000\n",
+	n_gets_diff = buf_pool->n_page_gets - buf_pool->n_page_gets_old;
+
+	if (n_gets_diff) {
+		fprintf(file,
+			"Buffer pool hit rate %lu / 1000,"
+			" young-making rate %lu / 1000 not %lu / 1000\n",
 			(ulong)
 			(1000 - ((1000 * (buf_pool->n_pages_read
 					  - buf_pool->n_pages_read_old))
-				 / (buf_pool->n_page_gets
-				    - buf_pool->n_page_gets_old))));
+				 / n_gets_diff)),
+			(ulong)
+			(1000 * (buf_pool->n_pages_made_young
+				 - buf_pool->n_pages_made_young_old)
+			 / n_gets_diff),
+			(ulong)
+			(1000 * (buf_pool->n_pages_not_made_young
+				 - buf_pool->n_pages_not_made_young_old)
+			 / n_gets_diff));
 	} else {
 		fputs("No buffer pool page gets since the last printout\n",
 		      file);
 	}
 
 	buf_pool->n_page_gets_old = buf_pool->n_page_gets;
+	buf_pool->n_pages_made_young_old = buf_pool->n_pages_made_young;
+	buf_pool->n_pages_not_made_young_old
+		= buf_pool->n_pages_not_made_young;
 	buf_pool->n_pages_read_old = buf_pool->n_pages_read;
 	buf_pool->n_pages_created_old = buf_pool->n_pages_created;
 	buf_pool->n_pages_written_old = buf_pool->n_pages_written;

buf/buf0rea.c

@@ -164,165 +164,6 @@ buf_read_page_low(
 	return(1);
 }
 
-/********************************************************************//**
-Applies a random read-ahead in buf_pool if there are at least a threshold
-value of accessed pages from the random read-ahead area. Does not read any
-page, not even the one at the position (space, offset), if the read-ahead
-mechanism is not activated. NOTE 1: the calling thread may own latches on
-pages: to avoid deadlocks this function must be written such that it cannot
-end up waiting for these latches! NOTE 2: the calling thread must want
-access to the page given: this rule is set to prevent unintended read-aheads
-performed by ibuf routines, a situation which could result in a deadlock if
-the OS does not support asynchronous i/o.
-@return number of page read requests issued; NOTE that if we read ibuf
-pages, it may happen that the page at the given page number does not
-get read even if we return a positive value! */
-static
-ulint
-buf_read_ahead_random(
-/*==================*/
-	ulint	space,	/*!< in: space id */
-	ulint	zip_size,/*!< in: compressed page size in bytes, or 0 */
-	ulint	offset)	/*!< in: page number of a page which the current thread
-			wants to access */
-{
-	ib_int64_t	tablespace_version;
-	ulint		recent_blocks	= 0;
-	ulint		count;
-	ulint		LRU_recent_limit;
-	ulint		ibuf_mode;
-	ulint		low, high;
-	ulint		err;
-	ulint		i;
-	ulint		buf_read_ahead_random_area;
-
-	/* We have currently disabled random readahead */
-	return(0);
-
-	if (srv_startup_is_before_trx_rollback_phase) {
-		/* No read-ahead to avoid thread deadlocks */
-		return(0);
-	}
-
-	if (ibuf_bitmap_page(zip_size, offset)
-	    || trx_sys_hdr_page(space, offset)) {
-
-		/* If it is an ibuf bitmap page or trx sys hdr, we do
-		no read-ahead, as that could break the ibuf page access
-		order */
-
-		return(0);
-	}
-
-	/* Remember the tablespace version before we ask te tablespace size
-	below: if DISCARD + IMPORT changes the actual .ibd file meanwhile, we
-	do not try to read outside the bounds of the tablespace! */
-
-	tablespace_version = fil_space_get_version(space);
-
-	buf_read_ahead_random_area = BUF_READ_AHEAD_RANDOM_AREA;
-
-	low  = (offset / buf_read_ahead_random_area)
-		* buf_read_ahead_random_area;
-	high = (offset / buf_read_ahead_random_area + 1)
-		* buf_read_ahead_random_area;
-	if (high > fil_space_get_size(space)) {
-
-		high = fil_space_get_size(space);
-	}
-
-	/* Get the minimum LRU_position field value for an initial segment
-	of the LRU list, to determine which blocks have recently been added
-	to the start of the list. */
-
-	LRU_recent_limit = buf_LRU_get_recent_limit();
-
-	buf_pool_mutex_enter();
-
-	if (buf_pool->n_pend_reads
-	    > buf_pool->curr_size / BUF_READ_AHEAD_PEND_LIMIT) {
-		buf_pool_mutex_exit();
-
-		return(0);
-	}
-
-	/* Count how many blocks in the area have been recently accessed,
-	that is, reside near the start of the LRU list. */
-
-	for (i = low; i < high; i++) {
-		const buf_page_t*	bpage = buf_page_hash_get(space, i);
-
-		if (bpage
-		    && buf_page_is_accessed(bpage)
-		    && (buf_page_get_LRU_position(bpage) > LRU_recent_limit)) {
-
-			recent_blocks++;
-
-			if (recent_blocks >= BUF_READ_AHEAD_RANDOM_THRESHOLD) {
-
-				buf_pool_mutex_exit();
-				goto read_ahead;
-			}
-		}
-	}
-
-	buf_pool_mutex_exit();
-	/* Do nothing */
-	return(0);
-
-read_ahead:
-	/* Read all the suitable blocks within the area */
-
-	if (ibuf_inside()) {
-		ibuf_mode = BUF_READ_IBUF_PAGES_ONLY;
-	} else {
-		ibuf_mode = BUF_READ_ANY_PAGE;
-	}
-
-	count = 0;
-
-	for (i = low; i < high; i++) {
-		/* It is only sensible to do read-ahead in the non-sync aio
-		mode: hence FALSE as the first parameter */
-
-		if (!ibuf_bitmap_page(zip_size, i)) {
-			count += buf_read_page_low(
-				&err, FALSE,
-				ibuf_mode | OS_AIO_SIMULATED_WAKE_LATER,
-				space, zip_size, FALSE,
-				tablespace_version, i);
-			if (err == DB_TABLESPACE_DELETED) {
-				ut_print_timestamp(stderr);
-				fprintf(stderr,
-					"  InnoDB: Warning: in random"
-					" readahead trying to access\n"
-					"InnoDB: tablespace %lu page %lu,\n"
-					"InnoDB: but the tablespace does not"
-					" exist or is just being dropped.\n",
-					(ulong) space, (ulong) i);
-			}
-		}
-	}
-
-	/* In simulated aio we wake the aio handler threads only after
-	queuing all aio requests, in native aio the following call does
-	nothing: */
-
-	os_aio_simulated_wake_handler_threads();
-
-#ifdef UNIV_DEBUG
-	if (buf_debug_prints && (count > 0)) {
-		fprintf(stderr,
-			"Random read-ahead space %lu offset %lu pages %lu\n",
-			(ulong) space, (ulong) offset,
-			(ulong) count);
-	}
-#endif /* UNIV_DEBUG */
-
-	++srv_read_ahead_rnd;
-	return(count);
-}
-
 /********************************************************************//**
 High-level function which reads a page asynchronously from a file to the
 buffer buf_pool if it is not already there. Sets the io_fix flag and sets
@@ -341,20 +182,17 @@ buf_read_page(
 {
 	ib_int64_t	tablespace_version;
 	ulint		count;
-	ulint		count2;
 	ulint		err;
 
 	tablespace_version = fil_space_get_version(space);
 
-	count = buf_read_ahead_random(space, zip_size, offset);
-
 	/* We do the i/o in the synchronous aio mode to save thread
 	switches: hence TRUE */
 
-	count2 = buf_read_page_low(&err, TRUE, BUF_READ_ANY_PAGE, space,
-				   zip_size, FALSE,
-				   tablespace_version, offset);
-	srv_buf_pool_reads+= count2;
+	count = buf_read_page_low(&err, TRUE, BUF_READ_ANY_PAGE, space,
+				  zip_size, FALSE,
+				  tablespace_version, offset);
+	srv_buf_pool_reads += count;
 	if (err == DB_TABLESPACE_DELETED) {
 		ut_print_timestamp(stderr);
 		fprintf(stderr,
@@ -371,7 +209,7 @@ buf_read_page(
 	/* Increment number of I/O operations used for LRU policy. */
 	buf_LRU_stat_inc_io();
 
-	return(count + count2);
+	return(count);
 }
 
 /********************************************************************//**
@@ -498,9 +336,17 @@ buf_read_ahead_linear(
 			fail_count++;
 
 		} else if (pred_bpage) {
-			int res = (ut_ulint_cmp(
-				       buf_page_get_LRU_position(bpage),
-				       buf_page_get_LRU_position(pred_bpage)));
+			/* Note that buf_page_is_accessed() returns
+			the time of the first access.  If some blocks
+			of the extent existed in the buffer pool at
+			the time of a linear access pattern, the first
+			access times may be nonmonotonic, even though
+			the latest access times were linear.  The
+			threshold (srv_read_ahead_factor) should help
+			a little against this. */
+			int res = ut_ulint_cmp(
+				buf_page_is_accessed(bpage),
+				buf_page_is_accessed(pred_bpage));
 			/* Accesses not in the right order */
 			if (res != 0 && res != asc_or_desc) {
 				fail_count++;

handler/ha_innodb.cc

@@ -72,6 +72,7 @@ with this program; if not, write to the Free Software Foundation, Inc.,
 /* Include necessary InnoDB headers */
 extern "C" {
 #include "univ.i"
+#include "buf0lru.h"
 #include "btr0sea.h"
 #include "os0file.h"
 #include "os0thread.h"
@@ -157,6 +158,10 @@ static ulong innobase_write_io_threads;
 
 static long long innobase_buffer_pool_size, innobase_log_file_size;
 
+/** Percentage of the buffer pool to reserve for 'old' blocks.
+Connected to buf_LRU_old_ratio. */
+static uint innobase_old_blocks_pct;
+
 /* The default values for the following char* start-up parameters
 are determined in innobase_init below: */
 
@@ -2208,6 +2213,9 @@ innobase_change_buffering_inited_ok:
 	ut_a(0 == strcmp(my_charset_latin1.name, "latin1_swedish_ci"));
 	srv_latin1_ordering = my_charset_latin1.sort_order;
 
+	innobase_old_blocks_pct = buf_LRU_old_ratio_update(
+		innobase_old_blocks_pct, FALSE);
+
 	innobase_commit_concurrency_init_default();
 
 	/* Since we in this module access directly the fields of a trx
@@ -9614,6 +9622,25 @@ innodb_adaptive_hash_index_update(
 	}
 }
 
+/****************************************************************//**
+Update the system variable innodb_old_blocks_pct using the "saved"
+value. This function is registered as a callback with MySQL. */
+static
+void
+innodb_old_blocks_pct_update(
+/*=========================*/
+	THD*				thd,	/*!< in: thread handle */
+	struct st_mysql_sys_var*	var,	/*!< in: pointer to
+						system variable */
+	void*				var_ptr,/*!< out: where the
+						formal string goes */
+	const void*			save)	/*!< in: immediate result
+						from check function */
+{
+	innobase_old_blocks_pct = buf_LRU_old_ratio_update(
+		*static_cast<const uint*>(save), TRUE);
+}
+
 /*************************************************************//**
 Check if it is a valid value of innodb_change_buffering.  This function is
 registered as a callback with MySQL.
@@ -9890,6 +9917,18 @@ static MYSQL_SYSVAR_LONG(mirrored_log_groups, innobase_mirrored_log_groups,
   "Number of identical copies of log groups we keep for the database. Currently this should be set to 1.",
   NULL, NULL, 1, 1, 10, 0);
 
+static MYSQL_SYSVAR_UINT(old_blocks_pct, innobase_old_blocks_pct,
+  PLUGIN_VAR_RQCMDARG,
+  "Percentage of the buffer pool to reserve for 'old' blocks.",
+  NULL, innodb_old_blocks_pct_update, 100 * 3 / 8, 5, 95, 0);
+
+static MYSQL_SYSVAR_UINT(old_blocks_time, buf_LRU_old_threshold_ms,
+  PLUGIN_VAR_RQCMDARG,
+  "Move blocks to the 'new' end of the buffer pool if the first access"
+  " was at least this many milliseconds ago."
+  " The timeout is disabled if 0 (the default).",
+  NULL, NULL, 0, 0, UINT_MAX32, 0);
+
 static MYSQL_SYSVAR_LONG(open_files, innobase_open_files,
   PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
   "How many files at the maximum InnoDB keeps open at the same time.",
@@ -9988,6 +10027,8 @@ static struct st_mysql_sys_var* innobase_system_variables[]= {
   MYSQL_SYSVAR(adaptive_flushing),
   MYSQL_SYSVAR(max_purge_lag),
   MYSQL_SYSVAR(mirrored_log_groups),
+  MYSQL_SYSVAR(old_blocks_pct),
+  MYSQL_SYSVAR(old_blocks_time),
   MYSQL_SYSVAR(open_files),
   MYSQL_SYSVAR(rollback_on_timeout),
   MYSQL_SYSVAR(stats_on_metadata),

include/buf0buf.h

@@ -707,15 +707,6 @@ buf_page_belongs_to_unzip_LRU(
 /*==========================*/
 	const buf_page_t*	bpage)	/*!< in: pointer to control block */
 	__attribute__((pure));
-/*********************************************************************//**
-Determine the approximate LRU list position of a block.
-@return	LRU list position */
-UNIV_INLINE
-ulint
-buf_page_get_LRU_position(
-/*======================*/
-	const buf_page_t*	bpage)	/*!< in: control block */
-	__attribute__((pure));
 
 /*********************************************************************//**
 Gets the mutex of a block.
@@ -816,22 +807,22 @@ buf_page_set_old(
 	buf_page_t*	bpage,	/*!< in/out: control block */
 	ibool		old);	/*!< in: old */
 /*********************************************************************//**
-Determine if a block has been accessed in the buffer pool.
-@return	TRUE if accessed */
+Determine the time of last access a block in the buffer pool.
+@return	ut_time_ms() at the time of last access, 0 if not accessed */
 UNIV_INLINE
-ibool
+unsigned
 buf_page_is_accessed(
 /*=================*/
 	const buf_page_t*	bpage)	/*!< in: control block */
-	__attribute__((pure));
+	__attribute__((nonnull, pure));
 /*********************************************************************//**
 Flag a block accessed. */
 UNIV_INLINE
 void
 buf_page_set_accessed(
 /*==================*/
-	buf_page_t*	bpage,		/*!< in/out: control block */
-	ibool		accessed);	/*!< in: accessed */
+	buf_page_t*	bpage)		/*!< in/out: control block */
+	__attribute__((nonnull));
 /*********************************************************************//**
 Gets the buf_block_t handle of a buffered file block if an uncompressed
 page frame exists, or NULL.
@@ -1017,14 +1008,6 @@ buf_block_hash_get(
 /*===============*/
 	ulint	space,	/*!< in: space id */
 	ulint	offset);/*!< in: offset of the page within space */
-/*******************************************************************//**
-Increments the pool clock by one and returns its new value. Remember that
-in the 32 bit version the clock wraps around at 4 billion!
-@return	new clock value */
-UNIV_INLINE
-ulint
-buf_pool_clock_tic(void);
-/*====================*/
 /*********************************************************************//**
 Gets the current length of the free list of buffer blocks.
 @return	length of the free list */
@@ -1064,16 +1047,10 @@ struct buf_page_struct{
 					flushed to disk, this tells the
 					flush_type.
 					@see enum buf_flush */
-	unsigned	accessed:1;	/*!< TRUE if the page has been accessed
-					while in the buffer pool: read-ahead
-					may read in pages which have not been
-					accessed yet; a thread is allowed to
-					read this for heuristic purposes
-					without holding any mutex or latch */
 	unsigned	io_fix:2;	/*!< type of pending I/O operation;
 					also protected by buf_pool_mutex
 					@see enum buf_io_fix */
-	unsigned	buf_fix_count:24;/*!< count of how manyfold this block
+	unsigned	buf_fix_count:25;/*!< count of how manyfold this block
 					is currently bufferfixed */
 	/* @} */
 #endif /* !UNIV_HOTBACKUP */
@@ -1152,17 +1129,7 @@ struct buf_page_struct{
 #endif /* UNIV_DEBUG */
 	unsigned	old:1;		/*!< TRUE if the block is in the old
 					blocks in the LRU list */
-	unsigned	LRU_position:31;/*!< value which monotonically
-					decreases (or may stay
-					constant if old==TRUE) toward
-					the end of the LRU list, if
-					buf_pool->ulint_clock has not
-					wrapped around: NOTE that this
-					value can only be used in
-					heuristic algorithms, because
-					of the possibility of a
-					wrap-around! */
-	unsigned	freed_page_clock:32;/*!< the value of
+	unsigned	freed_page_clock:31;/*!< the value of
 					buf_pool->freed_page_clock
 					when this block was the last
 					time put to the head of the
@@ -1170,6 +1137,9 @@ struct buf_page_struct{
 					to read this for heuristic
 					purposes without holding any
 					mutex or latch */
+	unsigned	access_time:32;	/*!< time of first access, or
+					0 if the block was never accessed
+					in the buffer pool */
 	/* @} */
 # ifdef UNIV_DEBUG_FILE_ACCESSES
 	ibool		file_page_was_freed;
@@ -1338,8 +1308,17 @@ struct buf_pool_struct{
 	ulint		n_pend_reads;	/*!< number of pending read operations */
 	ulint		n_pend_unzip;	/*!< number of pending decompressions */
 
-	time_t		last_printout_time; /*!< when buf_print was last time
+	time_t		last_printout_time;
+					/*!< when buf_print_io was last time
 					called */
+	ulint		n_pages_made_young;
+					/*!< number of pages made young, in
+					calls to buf_LRU_make_block_young() */
+	ulint		n_pages_not_made_young;
+					/*!< number of pages not made
+					young because the first access
+					was not long enough ago, in
+					buf_page_peek_if_too_old() */
 	ulint		n_pages_read;	/*!< number read operations */
 	ulint		n_pages_written;/*!< number write operations */
 	ulint		n_pages_created;/*!< number of pages created
@@ -1350,14 +1329,24 @@ struct buf_pool_struct{
 					counted as page gets; this field
 					is NOT protected by the buffer
 					pool mutex */
-	ulint		n_page_gets_old;/*!< n_page_gets when buf_print was
-					last time called: used to calculate
+	ulint		n_page_gets_old;/*!< n_page_gets when buf_print_io was
+					called last time: used to calculate
 					hit rate */
-	ulint		n_pages_read_old;/*!< n_pages_read when buf_print was
-					last time called */
-	ulint		n_pages_written_old;/*!< number write operations */
-	ulint		n_pages_created_old;/*!< number of pages created in
-					the pool with no read */
+	ulint		n_pages_made_young_old;
+					/*!< n_pages_made_young when
+					buf_print_io was called last time */
+	ulint		n_pages_not_made_young_old;
+					/*!< n_pages_not_made_young when
+					buf_print_io was called last time */
+	ulint		n_pages_read_old;
+					/*!< n_pages_read when buf_print_io
+					was called last time */
+	ulint		n_pages_written_old;
+					/*!< n_pages_written when buf_print_io
+					was called last time */
+	ulint		n_pages_created_old;
+					/*!< n_pages_created when buf_print_io
+					was called last time */
 	/* @} */
 	/** @name Page flushing algorithm fields */
 	/* @{ */
@@ -1375,10 +1364,6 @@ struct buf_pool_struct{
 					/*!< this is in the set state
 					when there is no flush batch
 					of the given type running */
-	ulint		ulint_clock;	/*!< a sequence number used to count
-					time. NOTE! This counter wraps
-					around at 4 billion (if ulint ==
-					32 bits)! */
 	ulint		freed_page_clock;/*!< a sequence number used
 					to count the number of buffer
 					blocks removed from the end of
@@ -1402,9 +1387,11 @@ struct buf_pool_struct{
 					block list */
 	UT_LIST_BASE_NODE_T(buf_page_t) LRU;
 					/*!< base node of the LRU list */
-	buf_page_t*	LRU_old;	/*!< pointer to the about 3/8 oldest
-					blocks in the LRU list; NULL if LRU
-					length less than BUF_LRU_OLD_MIN_LEN;
+	buf_page_t*	LRU_old;	/*!< pointer to the about
+					buf_LRU_old_ratio/BUF_LRU_OLD_RATIO_DIV
+					oldest blocks in the LRU list;
+					NULL if LRU length less than
+					BUF_LRU_OLD_MIN_LEN;
 					NOTE: when LRU_old != NULL, its length
 					should always equal LRU_old_len */
 	ulint		LRU_old_len;	/*!< length of the LRU list from

include/buf0buf.ic

@@ -72,9 +72,24 @@ buf_page_peek_if_too_old(
 /*=====================*/
 	const buf_page_t*	bpage)	/*!< in: block to make younger */
 {
-	return(buf_pool->freed_page_clock
-	       >= buf_page_get_freed_page_clock(bpage)
-	       + 1 + (buf_pool->curr_size / 4));
+	if (buf_LRU_old_threshold_ms && bpage->old) {
+		unsigned	access_time = buf_page_is_accessed(bpage);
+
+		if (access_time && ut_time_ms() - access_time
+		    >= buf_LRU_old_threshold_ms) {
+			return(TRUE);
+		}
+
+		buf_pool->n_pages_not_made_young++;
+		return(FALSE);
+	} else {
+		/* FIXME: bpage->freed_page_clock is 31 bits */
+		return((buf_pool->freed_page_clock & ~(1 << 31))
+		       > bpage->freed_page_clock
+		       + (buf_pool->curr_size
+			  * (BUF_LRU_OLD_RATIO_DIV - buf_LRU_old_ratio)
+			  / (BUF_LRU_OLD_RATIO_DIV * 4)));
+	}
 }
 
 /*********************************************************************//**
@@ -118,22 +133,6 @@ buf_pool_get_oldest_modification(void)
 
 	return(lsn);
 }
-
-/*******************************************************************//**
-Increments the buf_pool clock by one and returns its new value. Remember
-that in the 32 bit version the clock wraps around at 4 billion!
-@return	new clock value */
-UNIV_INLINE
-ulint
-buf_pool_clock_tic(void)
-/*====================*/
-{
-	ut_ad(buf_pool_mutex_own());
-
-	buf_pool->ulint_clock++;
-
-	return(buf_pool->ulint_clock);
-}
 #endif /* !UNIV_HOTBACKUP */
 
 /*********************************************************************//**
@@ -279,21 +278,6 @@ buf_page_belongs_to_unzip_LRU(
 	       && buf_page_get_state(bpage) == BUF_BLOCK_FILE_PAGE);
 }
 
-/*********************************************************************//**
-Determine the approximate LRU list position of a block.
-@return	LRU list position */
-UNIV_INLINE
-ulint
-buf_page_get_LRU_position(
-/*======================*/
-	const buf_page_t*	bpage)	/*!< in: control block */
-{
-	ut_ad(buf_page_in_file(bpage));
-	ut_ad(buf_pool_mutex_own());
-
-	return(bpage->LRU_position);
-}
-
 /*********************************************************************//**
 Gets the mutex of a block.
 @return	pointer to mutex protecting bpage */
@@ -487,17 +471,17 @@ buf_page_set_old(
 }
 
 /*********************************************************************//**
-Determine if a block has been accessed in the buffer pool.
-@return	TRUE if accessed */
+Determine the time of last access a block in the buffer pool.
+@return	ut_time_ms() at the time of last access, 0 if not accessed */
 UNIV_INLINE
-ibool
+unsigned
 buf_page_is_accessed(
 /*=================*/
 	const buf_page_t*	bpage)	/*!< in: control block */
 {
 	ut_ad(buf_page_in_file(bpage));
 
-	return(bpage->accessed);
+	return(bpage->access_time);
 }
 
 /*********************************************************************//**
@@ -506,13 +490,15 @@ UNIV_INLINE
 void
 buf_page_set_accessed(
 /*==================*/
-	buf_page_t*	bpage,		/*!< in/out: control block */
-	ibool		accessed)	/*!< in: accessed */
+	buf_page_t*	bpage)		/*!< in/out: control block */
 {
 	ut_a(buf_page_in_file(bpage));
-	ut_ad(mutex_own(buf_page_get_mutex(bpage)));
+	ut_ad(buf_pool_mutex_own());
 
-	bpage->accessed = accessed;
+	if (!bpage->access_time) {
+		/* Make this the time of the first access. */
+		bpage->access_time = ut_time_ms();
+	}
 }
 
 /*********************************************************************//**

include/buf0lru.h

@@ -69,7 +69,7 @@ These are low-level functions
 #########################################################################*/
 
 /** Minimum LRU list length for which the LRU_old pointer is defined */
-#define BUF_LRU_OLD_MIN_LEN	80
+#define BUF_LRU_OLD_MIN_LEN	512	/* 8 megabytes of 16k pages */
 
 /** Maximum LRU list search length in buf_flush_LRU_recommendation() */
 #define BUF_LRU_FREE_SEARCH_LEN		(5 + 2 * BUF_READ_AHEAD_AREA)
@@ -84,15 +84,6 @@ void
 buf_LRU_invalidate_tablespace(
 /*==========================*/
 	ulint	id);	/*!< in: space id */
-/******************************************************************//**
-Gets the minimum LRU_position field for the blocks in an initial segment
-(determined by BUF_LRU_INITIAL_RATIO) of the LRU list. The limit is not
-guaranteed to be precise, because the ulint_clock may wrap around.
-@return	the limit; zero if could not determine it */
-UNIV_INTERN
-ulint
-buf_LRU_get_recent_limit(void);
-/*==========================*/
 /********************************************************************//**
 Insert a compressed block into buf_pool->zip_clean in the LRU order. */
 UNIV_INTERN
@@ -201,6 +192,18 @@ void
 buf_LRU_make_block_old(
 /*===================*/
 	buf_page_t*	bpage);	/*!< in: control block */
+/**********************************************************************//**
+Updates buf_LRU_old_ratio.
+@return	updated old_pct */
+UNIV_INTERN
+uint
+buf_LRU_old_ratio_update(
+/*=====================*/
+	uint	old_pct,/*!< in: Reserve this percentage of
+			the buffer pool for "old" blocks. */
+	ibool	adjust);/*!< in: TRUE=adjust the LRU list;
+			FALSE=just assign buf_LRU_old_ratio
+			during the initialization of InnoDB */
 /********************************************************************//**
 Update the historical stats that we are collecting for LRU eviction
 policy at the end of each interval. */
@@ -227,6 +230,35 @@ buf_LRU_print(void);
 /*===============*/
 #endif /* UNIV_DEBUG_PRINT || UNIV_DEBUG || UNIV_BUF_DEBUG */
 
+/** @name Heuristics for detecting index scan @{ */
+/** Reserve this much/BUF_LRU_OLD_RATIO_DIV of the buffer pool for
+"old" blocks.  Protected by buf_pool_mutex. */
+extern uint	buf_LRU_old_ratio;
+/** The denominator of buf_LRU_old_ratio. */
+#define BUF_LRU_OLD_RATIO_DIV	1024
+/** Maximum value of buf_LRU_old_ratio.
+@see buf_LRU_old_adjust_len
+@see buf_LRU_old_ratio_update */
+#define BUF_LRU_OLD_RATIO_MAX	BUF_LRU_OLD_RATIO_DIV
+/** Minimum value of buf_LRU_old_ratio.
+@see buf_LRU_old_adjust_len
+@see buf_LRU_old_ratio_update
+The minimum must exceed
+(BUF_LRU_OLD_TOLERANCE + 5) * BUF_LRU_OLD_RATIO_DIV / BUF_LRU_OLD_MIN_LEN. */
+#define BUF_LRU_OLD_RATIO_MIN	51
+
+#if BUF_LRU_OLD_RATIO_MIN >= BUF_LRU_OLD_RATIO_MAX
+# error "BUF_LRU_OLD_RATIO_MIN >= BUF_LRU_OLD_RATIO_MAX"
+#endif
+#if BUF_LRU_OLD_RATIO_MAX > BUF_LRU_OLD_RATIO_DIV
+# error "BUF_LRU_OLD_RATIO_MAX > BUF_LRU_OLD_RATIO_DIV"
+#endif
+
+/** Move blocks to "new" LRU list only if the first access was at
+least this many milliseconds ago.  Not protected by any mutex or latch. */
+extern uint	buf_LRU_old_threshold_ms;
+/* @} */
+
 /** @brief Statistics for selecting the LRU list for eviction.
 
 These statistics are not 'of' LRU but 'for' LRU.  We keep count of I/O

include/ut0ut.h

@@ -239,6 +239,15 @@ ullint
 ut_time_us(
 /*=======*/
 	ullint*	tloc);	/*!< out: us since epoch, if non-NULL */
+/**********************************************************//**
+Returns the number of milliseconds since some epoch.  The
+value may wrap around.  It should only be used for heuristic
+purposes.
+@return	ms since epoch */
+UNIV_INTERN
+uint
+ut_time_ms(void);
+/*============*/
 
 /**********************************************************//**
 Returns the difference of two times in seconds.

ut/ut0ut.c

@@ -199,6 +199,23 @@ ut_time_us(
 	return(us);
 }
 
+/**********************************************************//**
+Returns the number of milliseconds since some epoch.  The
+value may wrap around.  It should only be used for heuristic
+purposes.
+@return	ms since epoch */
+UNIV_INTERN
+uint
+ut_time_ms(void)
+/*============*/
+{
+	struct timeval	tv;
+
+	ut_gettimeofday(&tv, NULL);
+
+	return((uint) tv.tv_sec * 1000 + tv.tv_usec / 1000);
+}
+
 /**********************************************************//**
 Returns the difference of two times in seconds.
 @return	time2 - time1 expressed in seconds */