Implement proposal for new scheduler wakeup strategy

6 files changed, 255 insertions, 47 deletions

diff --git a/erts/doc/src/erl.xml b/erts/doc/src/erl.xml
index 50f90ec9df..33a6339b04 100644
--- a/erts/doc/src/erl.xml
+++ b/erts/doc/src/erl.xml
@@ -867,6 +867,17 @@
 	    <p>For more information, see
 	    <seealso marker="erlang#system_flag_cpu_topology">erlang:system_flag(cpu_topology, CpuTopology)</seealso>.</p>
           </item>
+	  <tag><marker id="+sws"><c>+sws default|legacy|proposal</c></marker></tag>
+	  <item>
+             <p>Set scheduler wakeup strategy. Default is <c>legacy</c> (has been
+	        used since OTP-R13B). The <c>proposal</c> strategy is the currently
+		proposed strategy for OTP-R16. Note that the <c>proposal</c> strategy
+		might change during OTP-R15.
+	     </p>
+             <p><em>NOTE:</em> This flag may be removed or changed at any time
+	        without prior notice.
+	     </p>
+	  </item>
 	  <tag><marker id="+swt"><c>+swt very_low|low|medium|high|very_high</c></marker></tag>
 	  <item>
              <p>Set scheduler wakeup threshold. Default is <c>medium</c>.
diff --git a/erts/emulator/beam/erl_init.c b/erts/emulator/beam/erl_init.c
index 02efd2adb6..066ceff194 100644
--- a/erts/emulator/beam/erl_init.c
+++ b/erts/emulator/beam/erl_init.c
@@ -514,6 +514,8 @@ void erts_usage(void)
     erts_fprintf(stderr, "            see the erl(1) documentation for more info.\n");
     erts_fprintf(stderr, "-sct cput   set cpu topology,\n");
     erts_fprintf(stderr, "            see the erl(1) documentation for more info.\n");
+    erts_fprintf(stderr, "-sws val    set scheduler wakeup strategy, valid values are:\n");
+    erts_fprintf(stderr, "            default|legacy|proposal.\n");
     erts_fprintf(stderr, "-swt val    set scheduler wakeup threshold, valid values are:\n");
     erts_fprintf(stderr, "            very_low|low|medium|high|very_high.\n");
     erts_fprintf(stderr, "-sss size   suggested stack size in kilo words for scheduler threads,\n");
@@ -1217,13 +1219,23 @@ erl_start(int argc, char **argv)
 		use_multi_run_queue = 0;
 	    else if (sys_strcmp("wt", sub_param) == 0) {
 		arg = get_arg(sub_param+2, argv[i+1], &i);
-		if (erts_sched_set_wakeup_limit(arg) != 0) {
+		if (erts_sched_set_wakeup_other_thresold(arg) != 0) {
 		    erts_fprintf(stderr, "scheduler wakeup threshold: %s\n",
 				 arg);
 		    erts_usage();
 		}
 		VERBOSE(DEBUG_SYSTEM,
-			("scheduler wakup threshold: %s\n", arg));
+			("scheduler wakeup threshold: %s\n", arg));
+	    }
+	    else if (sys_strcmp("ws", sub_param) == 0) {
+		arg = get_arg(sub_param+2, argv[i+1], &i);
+		if (erts_sched_set_wakeup_other_type(arg) != 0) {
+		    erts_fprintf(stderr, "scheduler wakeup strategy: %s\n",
+				 arg);
+		    erts_usage();
+		}
+		VERBOSE(DEBUG_SYSTEM,
+			("scheduler wakeup threshold: %s\n", arg));
 	    }
 	    else if (has_prefix("ss", sub_param)) {
 		/* suggested stack size (Kilo Words) for scheduler threads */
diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c
index dba57e04ca..229e3c52dd 100644
--- a/erts/emulator/beam/erl_process.c
+++ b/erts/emulator/beam/erl_process.c
@@ -54,15 +54,6 @@
   (ERTS_SCHED_SYS_SLEEP_SPINCOUNT*ERTS_SCHED_TSE_SLEEP_SPINCOUNT_FACT)
 #define ERTS_SCHED_SUSPEND_SLEEP_SPINCOUNT 0
 
-#define ERTS_WAKEUP_OTHER_LIMIT_VERY_HIGH (200*CONTEXT_REDS)
-#define ERTS_WAKEUP_OTHER_LIMIT_HIGH (50*CONTEXT_REDS)
-#define ERTS_WAKEUP_OTHER_LIMIT_MEDIUM (10*CONTEXT_REDS)
-#define ERTS_WAKEUP_OTHER_LIMIT_LOW (CONTEXT_REDS)
-#define ERTS_WAKEUP_OTHER_LIMIT_VERY_LOW (CONTEXT_REDS/10)
-
-#define ERTS_WAKEUP_OTHER_DEC 10
-#define ERTS_WAKEUP_OTHER_FIXED_INC (CONTEXT_REDS/10)
-
 #if 0 || defined(DEBUG)
 #define ERTS_FAKE_SCHED_BIND_PRINT_SORTED_CPU_DATA
 #endif
@@ -116,8 +107,6 @@ Uint erts_no_schedulers;
 Uint erts_max_processes = ERTS_DEFAULT_MAX_PROCESSES;
 Uint erts_process_tab_index_mask;
 
-static int wakeup_other_limit;
-
 int erts_sched_thread_suggested_stack_size = -1;
 
 #ifdef ERTS_ENABLE_LOCK_CHECK
@@ -2554,30 +2543,243 @@ erts_debug_nbalance(void)
 #endif
 }
 
+/* Wakeup other schedulers */
+
+typedef enum {
+    ERTS_SCHED_WAKEUP_OTHER_THRESHOLD_VERY_HIGH,
+    ERTS_SCHED_WAKEUP_OTHER_THRESHOLD_HIGH,
+    ERTS_SCHED_WAKEUP_OTHER_THRESHOLD_MEDIUM,
+    ERTS_SCHED_WAKEUP_OTHER_THRESHOLD_LOW,
+    ERTS_SCHED_WAKEUP_OTHER_THRESHOLD_VERY_LOW
+} ErtsSchedWakeupOtherThreshold;
+
+typedef enum {
+    ERTS_SCHED_WAKEUP_OTHER_TYPE_PROPOSAL,
+    ERTS_SCHED_WAKEUP_OTHER_TYPE_LEGACY
+} ErtsSchedWakeupOtherType;
+
+/* First proposal */
+
+#define ERTS_WAKEUP_OTHER_LIMIT_VERY_HIGH (200*CONTEXT_REDS)
+#define ERTS_WAKEUP_OTHER_LIMIT_HIGH (50*CONTEXT_REDS)
+#define ERTS_WAKEUP_OTHER_LIMIT_MEDIUM (10*CONTEXT_REDS)
+#define ERTS_WAKEUP_OTHER_LIMIT_LOW (CONTEXT_REDS)
+#define ERTS_WAKEUP_OTHER_LIMIT_VERY_LOW (CONTEXT_REDS/10)
+
+#define ERTS_WAKEUP_OTHER_DEC_SHIFT_VERY_HIGH 3
+#define ERTS_WAKEUP_OTHER_DEC_SHIFT_HIGH 1
+#define ERTS_WAKEUP_OTHER_DEC_SHIFT_MEDIUM 0
+#define ERTS_WAKEUP_OTHER_DEC_SHIFT_LOW -2
+#define ERTS_WAKEUP_OTHER_DEC_SHIFT_VERY_LOW -5 
+
+#define ERTS_WAKEUP_OTHER_DEC_SHIFT 2
+#define ERTS_WAKEUP_OTHER_FIXED_INC (CONTEXT_REDS/10)
+
+/* To be legacy */
+
+#define ERTS_WAKEUP_OTHER_LIMIT_VERY_HIGH_LEGACY (200*CONTEXT_REDS)
+#define ERTS_WAKEUP_OTHER_LIMIT_HIGH_LEGACY (50*CONTEXT_REDS)
+#define ERTS_WAKEUP_OTHER_LIMIT_MEDIUM_LEGACY (10*CONTEXT_REDS)
+#define ERTS_WAKEUP_OTHER_LIMIT_LOW_LEGACY (CONTEXT_REDS)
+#define ERTS_WAKEUP_OTHER_LIMIT_VERY_LOW_LEGACY (CONTEXT_REDS/10)
+
+#define ERTS_WAKEUP_OTHER_DEC_LEGACY 10
+#define ERTS_WAKEUP_OTHER_FIXED_INC_LEGACY (CONTEXT_REDS/10)
+
+#ifdef ERTS_SMP
+
+static struct {
+    ErtsSchedWakeupOtherThreshold threshold;
+    ErtsSchedWakeupOtherType type;
+    int limit;
+    int dec_shift;
+    int dec_mask;
+    void (*check)(ErtsRunQueue *rq);
+} wakeup_other;
+
+static void
+wakeup_other_check(ErtsRunQueue *rq)
+{
+    int wo_reds = rq->wakeup_other_reds;
+    if (wo_reds) {
+	int left_len = rq->len - 1;
+	if (left_len < 1) {
+	    int wo_reduce = wo_reds << wakeup_other.dec_shift;
+	    wo_reduce &= wakeup_other.dec_mask;
+	    rq->wakeup_other -= wo_reduce;
+	    if (rq->wakeup_other < 0)
+		rq->wakeup_other = 0;
+	}
+	else {
+	    rq->wakeup_other += (left_len*wo_reds
+				 + ERTS_WAKEUP_OTHER_FIXED_INC);
+	    if (rq->wakeup_other > wakeup_other.limit) {
+		if (erts_common_run_queue) {
+		    if (erts_common_run_queue->waiting)
+			wake_scheduler(erts_common_run_queue, 0, 1);
+		}
+		else {
+		    int empty_rqs =
+			erts_smp_atomic32_read_acqb(&no_empty_run_queues);
+		    if (empty_rqs != 0)
+			wake_scheduler_on_empty_runq(rq);
+		}
+		rq->wakeup_other = 0;
+	    }
+	}
+	rq->wakeup_other_reds = 0;
+    }
+}
+
+static void
+wakeup_other_set_limit(void)
+{
+    switch (wakeup_other.threshold) {
+    case ERTS_SCHED_WAKEUP_OTHER_THRESHOLD_VERY_HIGH:
+	wakeup_other.limit = ERTS_WAKEUP_OTHER_LIMIT_VERY_HIGH;
+	wakeup_other.dec_shift = ERTS_WAKEUP_OTHER_DEC_SHIFT_VERY_HIGH;
+	break;
+    case ERTS_SCHED_WAKEUP_OTHER_THRESHOLD_HIGH:
+	wakeup_other.limit = ERTS_WAKEUP_OTHER_LIMIT_HIGH;
+	wakeup_other.dec_shift = ERTS_WAKEUP_OTHER_DEC_SHIFT_HIGH;
+	break;
+    case ERTS_SCHED_WAKEUP_OTHER_THRESHOLD_MEDIUM:
+	wakeup_other.limit = ERTS_WAKEUP_OTHER_LIMIT_MEDIUM;
+	wakeup_other.dec_shift = ERTS_WAKEUP_OTHER_DEC_SHIFT_MEDIUM;
+	break;
+    case ERTS_SCHED_WAKEUP_OTHER_THRESHOLD_LOW:
+	wakeup_other.limit = ERTS_WAKEUP_OTHER_LIMIT_LOW;
+	wakeup_other.dec_shift = ERTS_WAKEUP_OTHER_DEC_SHIFT_LOW;
+	break;
+    case ERTS_SCHED_WAKEUP_OTHER_THRESHOLD_VERY_LOW:
+	wakeup_other.limit = ERTS_WAKEUP_OTHER_LIMIT_VERY_LOW;
+	wakeup_other.dec_shift = ERTS_WAKEUP_OTHER_DEC_SHIFT_VERY_LOW;
+	break;
+    }
+    if (wakeup_other.dec_shift < 0)
+	wakeup_other.dec_mask = (1 << (sizeof(wakeup_other.dec_mask)*8
+				       + wakeup_other.dec_shift)) - 1;
+    else {
+	wakeup_other.dec_mask = 0;
+	wakeup_other.dec_mask = ~wakeup_other.dec_mask;
+    }
+}
+
+static void
+wakeup_other_check_legacy(ErtsRunQueue *rq)
+{
+    int wo_reds = rq->wakeup_other_reds;
+    if (wo_reds) {
+	if (rq->len < 2) {
+	    rq->wakeup_other -= ERTS_WAKEUP_OTHER_DEC_LEGACY*wo_reds;
+	    if (rq->wakeup_other < 0)
+		rq->wakeup_other = 0;
+	}
+	else if (rq->wakeup_other < wakeup_other.limit)
+	    rq->wakeup_other += rq->len*wo_reds + ERTS_WAKEUP_OTHER_FIXED_INC_LEGACY;
+	else {
+	    if (erts_common_run_queue) {
+		if (erts_common_run_queue->waiting)
+		    wake_scheduler(erts_common_run_queue, 0, 1);
+	    }
+	    else if (erts_smp_atomic32_read_acqb(&no_empty_run_queues) != 0) {
+		wake_scheduler_on_empty_runq(rq);
+		rq->wakeup_other = 0;
+	    }
+	    rq->wakeup_other = 0;
+	}
+    }
+    rq->wakeup_other_reds = 0;
+}
+
+static void
+wakeup_other_set_limit_legacy(void)
+{
+    switch (wakeup_other.threshold) {
+    case ERTS_SCHED_WAKEUP_OTHER_THRESHOLD_VERY_HIGH:
+	wakeup_other.limit = ERTS_WAKEUP_OTHER_LIMIT_VERY_HIGH_LEGACY;
+	break;
+    case ERTS_SCHED_WAKEUP_OTHER_THRESHOLD_HIGH:
+	wakeup_other.limit = ERTS_WAKEUP_OTHER_LIMIT_HIGH_LEGACY;
+	break;
+    case ERTS_SCHED_WAKEUP_OTHER_THRESHOLD_MEDIUM:
+	wakeup_other.limit = ERTS_WAKEUP_OTHER_LIMIT_MEDIUM_LEGACY;
+	break;
+    case ERTS_SCHED_WAKEUP_OTHER_THRESHOLD_LOW:
+	wakeup_other.limit = ERTS_WAKEUP_OTHER_LIMIT_LOW_LEGACY;
+	break;
+    case ERTS_SCHED_WAKEUP_OTHER_THRESHOLD_VERY_LOW:
+	wakeup_other.limit = ERTS_WAKEUP_OTHER_LIMIT_VERY_LOW_LEGACY;
+	break;
+    }
+}
+
+static void
+set_wakeup_other_data(void)
+{
+    switch (wakeup_other.type) {
+    case ERTS_SCHED_WAKEUP_OTHER_TYPE_PROPOSAL:
+	wakeup_other.check = wakeup_other_check;
+	wakeup_other_set_limit();
+	break;
+    case ERTS_SCHED_WAKEUP_OTHER_TYPE_LEGACY:
+	wakeup_other.check = wakeup_other_check_legacy;
+	wakeup_other_set_limit_legacy();
+	break;
+    }
+}
+
+#endif
+
 void
 erts_early_init_scheduling(void)
 {
-    wakeup_other_limit = ERTS_WAKEUP_OTHER_LIMIT_MEDIUM;
+#ifdef ERTS_SMP
+    wakeup_other.threshold = ERTS_SCHED_WAKEUP_OTHER_THRESHOLD_MEDIUM;
+    wakeup_other.type = ERTS_SCHED_WAKEUP_OTHER_TYPE_LEGACY;
+#endif
 }
 
 int
-erts_sched_set_wakeup_limit(char *str)
+erts_sched_set_wakeup_other_thresold(char *str)
 {
+    ErtsSchedWakeupOtherThreshold threshold;
     if (sys_strcmp(str, "very_high") == 0)
-	wakeup_other_limit = ERTS_WAKEUP_OTHER_LIMIT_VERY_HIGH;
+	threshold = ERTS_SCHED_WAKEUP_OTHER_THRESHOLD_VERY_HIGH;
     else if (sys_strcmp(str, "high") == 0)
-	wakeup_other_limit = ERTS_WAKEUP_OTHER_LIMIT_HIGH;
+	threshold = ERTS_SCHED_WAKEUP_OTHER_THRESHOLD_HIGH;
     else if (sys_strcmp(str, "medium") == 0)
-	wakeup_other_limit = ERTS_WAKEUP_OTHER_LIMIT_MEDIUM;
+	threshold = ERTS_SCHED_WAKEUP_OTHER_THRESHOLD_MEDIUM;
     else if (sys_strcmp(str, "low") == 0)
-	wakeup_other_limit = ERTS_WAKEUP_OTHER_LIMIT_LOW;
+	threshold = ERTS_SCHED_WAKEUP_OTHER_THRESHOLD_LOW;
     else if (sys_strcmp(str, "very_low") == 0)
-	wakeup_other_limit = ERTS_WAKEUP_OTHER_LIMIT_VERY_LOW;
+	threshold = ERTS_SCHED_WAKEUP_OTHER_THRESHOLD_VERY_LOW;
     else
 	return EINVAL;
+#ifdef ERTS_SMP
+    wakeup_other.threshold = threshold;
+    set_wakeup_other_data();
+#endif
+    return 0;
+}
+
+int
+erts_sched_set_wakeup_other_type(char *str)
+{
+    ErtsSchedWakeupOtherType type;
+    if (sys_strcmp(str, "proposal") == 0)
+	type = ERTS_SCHED_WAKEUP_OTHER_TYPE_PROPOSAL;
+    else if (sys_strcmp(str, "default") == 0)
+	type = ERTS_SCHED_WAKEUP_OTHER_TYPE_LEGACY;
+    else if (sys_strcmp(str, "legacy") == 0)
+	type = ERTS_SCHED_WAKEUP_OTHER_TYPE_LEGACY;
+    else
+	return EINVAL;
+#ifdef ERTS_SMP
+    wakeup_other.type = type;
+#endif
     return 0;
 }
-	
 
 void
 erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online)
@@ -2590,6 +2792,10 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online)
 
     init_misc_op_list_alloc();
 
+#ifdef ERTS_SMP
+    set_wakeup_other_data();
+#endif
+
     ASSERT(no_schedulers_online <= no_schedulers);
     ASSERT(no_schedulers_online >= 1);
     ASSERT(no_schedulers >= 1);
@@ -5480,30 +5686,7 @@ Process *schedule(Process *p, int calls)
 	    exec_misc_ops(rq);
 
 #ifdef ERTS_SMP
-	{
-	    int wo_reds = rq->wakeup_other_reds;
-	    if (wo_reds) {
-		if (rq->len < 2) {
-		    rq->wakeup_other -= ERTS_WAKEUP_OTHER_DEC*wo_reds;
-		    if (rq->wakeup_other < 0)
-			rq->wakeup_other = 0;
-		}
-		else if (rq->wakeup_other < wakeup_other_limit)
-		    rq->wakeup_other += rq->len*wo_reds + ERTS_WAKEUP_OTHER_FIXED_INC;
-		else {
-		    if (erts_common_run_queue) {
-			if (erts_common_run_queue->waiting)
-			    wake_scheduler(erts_common_run_queue, 0, 1);
-		    }
-		    else if (erts_smp_atomic32_read_acqb(&no_empty_run_queues) != 0) {
-			wake_scheduler_on_empty_runq(rq);
-			rq->wakeup_other = 0;
-		    }
-		    rq->wakeup_other = 0;
-		}
-	    }
-	    rq->wakeup_other_reds = 0;
-	}
+	wakeup_other.check(rq);
 #endif
 
 	/*
diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h
index c7d89d3bb3..d2f0c67597 100644
--- a/erts/emulator/beam/erl_process.h
+++ b/erts/emulator/beam/erl_process.h
@@ -1040,7 +1040,8 @@ ErtsProcList *erts_proclist_create(Process *);
 void erts_proclist_destroy(ErtsProcList *);
 int erts_proclist_same(ErtsProcList *, Process *);
 
-int erts_sched_set_wakeup_limit(char *str);
+int erts_sched_set_wakeup_other_thresold(char *str);
+int erts_sched_set_wakeup_other_type(char *str);
 
 #ifdef DEBUG
 void erts_dbg_multi_scheduling_return_trap(Process *, Eterm);
diff --git a/erts/etc/common/erlexec.c b/erts/etc/common/erlexec.c
index 95cb688b29..59bc7229e4 100644
--- a/erts/etc/common/erlexec.c
+++ b/erts/etc/common/erlexec.c
@@ -125,6 +125,7 @@ static char *pluss_val_switches[] = {
     "cl",
     "ct",
     "wt",
+    "ws",
     "ss",
     NULL
 };
diff --git a/erts/vsn.mk b/erts/vsn.mk
index 6be703d453..5af5db3c94 100644
--- a/erts/vsn.mk
+++ b/erts/vsn.mk
@@ -17,7 +17,7 @@
 # %CopyrightEnd%
 # 
 
-VSN = 5.8.4.1
+VSN = 5.8.4.1.0.2
 SYSTEM_VSN = R14B03
 
 # Port number 4365 in 4.2