branches/zip:

Fix Mantis Issue#244 fix bug in linear read ahead (no check on access pattern)

The changes are:

1) Take into account access pattern when deciding whether or not to do linear
  read ahead.
2) Expose a knob innodb_read_ahead_factor = [0-64] default (8), dynamic,
  global to control linear read ahead behvior
3) Disable random read ahead. Keep the code for now.

Submitted by:	Inaam (rb://122)
Approved by:	Heikki (rb://122)

buf/buf0lru.c

@@ -456,11 +456,12 @@ buf_LRU_get_recent_limit(void)
 
 	bpage = UT_LIST_GET_FIRST(buf_pool->LRU);
 
-	limit = buf_page_get_LRU_position(bpage) - len / BUF_LRU_INITIAL_RATIO;
+	limit = buf_page_get_LRU_position(bpage);
+	len /= BUF_LRU_INITIAL_RATIO;
 
 	buf_pool_mutex_exit();
 
-	return(limit);
+	return(limit > len ? (limit - len) : 0);
 }
 
 /********************************************************************//**

buf/buf0rea.c

@@ -44,14 +44,11 @@ the accessed pages when deciding whether to read-ahead */
 
 /** There must be at least this many pages in buf_pool in the area to start
 a random read-ahead */
-#define BUF_READ_AHEAD_RANDOM_THRESHOLD	(5 + buf_read_ahead_random_area / 8)
+#define BUF_READ_AHEAD_RANDOM_THRESHOLD	(1 + BUF_READ_AHEAD_RANDOM_AREA / 2)
 
 /** The linear read-ahead area size */
 #define	BUF_READ_AHEAD_LINEAR_AREA	BUF_READ_AHEAD_AREA
 
-/** The linear read-ahead threshold */
-#define LINEAR_AREA_THRESHOLD_COEF	5 / 8
-
 /** If there are buf_pool->curr_size per the number below pending reads, then
 read-ahead is not done: this is to prevent flooding the buffer pool with
 i/o-fixed buffer blocks */
@@ -199,6 +196,9 @@ buf_read_ahead_random(
 	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);
@@ -423,6 +423,7 @@ buf_read_ahead_linear(
 	ulint		i;
 	const ulint	buf_read_ahead_linear_area
 		= BUF_READ_AHEAD_LINEAR_AREA;
+	ulint		threshold;
 
 	if (UNIV_UNLIKELY(srv_startup_is_before_trx_rollback_phase)) {
 		/* No read-ahead to avoid thread deadlocks */
@@ -482,6 +483,10 @@ buf_read_ahead_linear(
 		asc_or_desc = -1;
 	}
 
+	/* How many out of order accessed pages can we ignore
+	when working out the access pattern for linear readahead */
+	threshold = ut_min(srv_read_ahead_factor, BUF_READ_AHEAD_AREA);
+
 	fail_count = 0;
 
 	for (i = low; i < high; i++) {
@@ -491,25 +496,25 @@ buf_read_ahead_linear(
 			/* Not accessed */
 			fail_count++;
 
-		} else if (pred_bpage
-			   && (ut_ulint_cmp(
+		} else if (pred_bpage) {
+			int res = (ut_ulint_cmp(
 				       buf_page_get_LRU_position(bpage),
-				       buf_page_get_LRU_position(pred_bpage))
-			       != asc_or_desc)) {
+				       buf_page_get_LRU_position(pred_bpage)));
 			/* Accesses not in the right order */
-
-			fail_count++;
-			pred_bpage = bpage;
+			if (res != 0 && res != asc_or_desc) {
+				fail_count++;
+			}
 		}
-	}
-
-	if (fail_count > buf_read_ahead_linear_area
-	    * LINEAR_AREA_THRESHOLD_COEF) {
-		/* Too many failures: return */
 
-		buf_pool_mutex_exit();
+		if (fail_count > threshold) {
+			/* Too many failures: return */
+			buf_pool_mutex_exit();
+			return(0);
+		}
 
-		return(0);
+		if (bpage && buf_page_is_accessed(bpage)) {
+			pred_bpage = bpage;
+		}
 	}
 
 	/* If we got this far, we know that enough pages in the area have

handler/ha_innodb.cc

@@ -9735,6 +9735,14 @@ static MYSQL_SYSVAR_STR(change_buffering, innobase_change_buffering,
   innodb_change_buffering_validate,
   innodb_change_buffering_update, NULL);
 
+static MYSQL_SYSVAR_ULONG(read_ahead_factor, srv_read_ahead_factor,
+  PLUGIN_VAR_RQCMDARG,
+  "Number of pages that may be accessed out of order and InnoDB "
+  "will still issue a readahead."
+  "The higher the value the more relaxed the condition for starting "
+  "readahead.",
+  NULL, NULL, 8, 0, 64, 0);
+
 static struct st_mysql_sys_var* innobase_system_variables[]= {
   MYSQL_SYSVAR(additional_mem_pool_size),
   MYSQL_SYSVAR(autoextend_increment),
@@ -9784,6 +9792,7 @@ static struct st_mysql_sys_var* innobase_system_variables[]= {
   MYSQL_SYSVAR(version),
   MYSQL_SYSVAR(use_sys_malloc),
   MYSQL_SYSVAR(change_buffering),
+  MYSQL_SYSVAR(read_ahead_factor),
   NULL
 };
 

include/srv0srv.h

@@ -125,6 +125,7 @@ extern ulint	srv_mem_pool_size;
 extern ulint	srv_lock_table_size;
 
 extern ulint	srv_n_file_io_threads;
+extern ulint	srv_read_ahead_factor;
 
 #ifdef UNIV_LOG_ARCHIVE
 extern ibool	srv_log_archive_on;

srv/srv0srv.c

@@ -175,6 +175,13 @@ UNIV_INTERN ulint	srv_lock_table_size	= ULINT_MAX;
 
 UNIV_INTERN ulint	srv_n_file_io_threads	= ULINT_MAX;
 
+/* User settable value of the number of pages that InnoDB will tolerate
+within a 64 page extent even if they are accessed out of order or have
+not been accessed at all. This number (which varies from 0 to 64) is
+indicative of the slack that we have when deciding about linear
+readahead. */
+UNIV_INTERN ulint	srv_read_ahead_factor	= 8;
+
 #ifdef UNIV_LOG_ARCHIVE
 UNIV_INTERN ibool		srv_log_archive_on	= FALSE;
 UNIV_INTERN ibool		srv_archive_recovery	= 0;