[PATCH] xen/sched_credit: Use delay to control scheduling frequency

[Hui Lv <hui.lv@intel.com>]

1. Basically, the "delay method" can achieve 11% overall performance boost for SPECvirt than original credit scheduler.
2. We have tried 1ms delay and 10ms delay, there is no big difference between these two configurations. (1ms is enough to achieve a good performance)
3. We have compared different load level response time/latency (low, high, peak), "delay method" didn't bring very much response time increase.
4. 1ms delay can reduce 30% context switch at peak performance, where produces the benefits. (.int sched_ratelimit_us = 1000. is the recommended setting)

Signed-off-by: Hui Lv <hui.lv@intel.com>
Signed-off-by: George Dunlap <George.Dunlap@eu.citrix.com>

diff -r a4bff36780a3 -r 114495d1ff94 xen/common/sched_credit.c
--- a/xen/common/sched_credit.c	Fri Dec 16 18:46:27 2011 +0000
+++ b/xen/common/sched_credit.c	Mon Dec 19 15:58:50 2011 -0500
@@ -110,6 +110,10 @@ boolean_param("sched_credit_default_yiel
 static int __read_mostly sched_credit_tslice_ms = CSCHED_DEFAULT_TSLICE_MS;
 integer_param("sched_credit_tslice_ms", sched_credit_tslice_ms);
 
+/* Scheduler generic parameters
+*/
+extern int sched_ratelimit_us;
+
 /*
  * Physical CPU
  */
@@ -1297,10 +1301,15 @@ csched_schedule(
     struct csched_private *prv = CSCHED_PRIV(ops);
     struct csched_vcpu *snext;
     struct task_slice ret;
+    s_time_t runtime, tslice;
 
     CSCHED_STAT_CRANK(schedule);
     CSCHED_VCPU_CHECK(current);
 
+    runtime = now - current->runstate.state_entry_time;
+    if ( runtime < 0 ) /* Does this ever happen? */
+        runtime = 0;
+
     if ( !is_idle_vcpu(scurr->vcpu) )
     {
         /* Update credits of a non-idle VCPU. */
@@ -1313,6 +1322,41 @@ csched_schedule(
         scurr->pri = CSCHED_PRI_IDLE;
     }
 
+    /* Choices, choices:
+     * - If we have a tasklet, we need to run the idle vcpu no matter what.
+     * - If sched rate limiting is in effect, and the current vcpu has
+     *   run for less than that amount of time, continue the current one,
+     *   but with a shorter timeslice and return it immediately
+     * - Otherwise, chose the one with the highest priority (which may
+     *   be the one currently running)
+     * - If the currently running one is TS_OVER, see if there
+     *   is a higher priority one waiting on the runqueue of another
+     *   cpu and steal it.
+     */
+
+    /* If we have schedule rate limiting enabled, check to see
+     * how long we've run for. */
+    if ( !tasklet_work_scheduled
+         && sched_ratelimit_us
+         && vcpu_runnable(current)
+         && !is_idle_vcpu(current)
+         && runtime < MICROSECS(sched_ratelimit_us) )
+    {
+        snext = scurr;
+        snext->start_time += now;
+        perfc_incr(delay_ms);
+        tslice = MICROSECS(sched_ratelimit_us);
+        ret.migrated = 0;
+        goto out;
+    }
+    else
+    {
+        /*
+         * Select next runnable local VCPU (ie top of local runq)
+        */
+        tslice = MILLISECS(prv->tslice_ms);
+    }
+
     /*
      * Select next runnable local VCPU (ie top of local runq)
      */
@@ -1367,11 +1411,12 @@ csched_schedule(
     if ( !is_idle_vcpu(snext->vcpu) )
         snext->start_time += now;
 
+out:
     /*
      * Return task to run next...
      */
     ret.time = (is_idle_vcpu(snext->vcpu) ?
-                -1 : MILLISECS(prv->tslice_ms));
+                -1 : tslice);
     ret.task = snext->vcpu;
 
     CSCHED_VCPU_CHECK(ret.task);
diff -r a4bff36780a3 -r 114495d1ff94 xen/common/schedule.c
--- a/xen/common/schedule.c	Fri Dec 16 18:46:27 2011 +0000
+++ b/xen/common/schedule.c	Mon Dec 19 15:58:50 2011 -0500
@@ -47,6 +47,12 @@ string_param("sched", opt_sched);
 bool_t sched_smt_power_savings = 0;
 boolean_param("sched_smt_power_savings", sched_smt_power_savings);
 
+/* Default scheduling rate limit: 1ms 
+ * It's not recommended to set this value bigger than "sched_credit_tslice_ms" 
+ * otherwise, something weired may happen
+ * */
+int sched_ratelimit_us = 1000;
+integer_param("sched_ratelimit_us", sched_ratelimit_us);
 /* Various timer handlers. */
 static void s_timer_fn(void *unused);
 static void vcpu_periodic_timer_fn(void *data);
diff -r a4bff36780a3 -r 114495d1ff94 xen/include/xen/perfc_defn.h
--- a/xen/include/xen/perfc_defn.h	Fri Dec 16 18:46:27 2011 +0000
+++ b/xen/include/xen/perfc_defn.h	Mon Dec 19 15:58:50 2011 -0500
@@ -16,6 +16,7 @@ PERFCOUNTER(sched_irq,              "sch
 PERFCOUNTER(sched_run,              "sched: runs through scheduler")
 PERFCOUNTER(sched_ctx,              "sched: context switches")
 
+PERFCOUNTER(delay_ms,               "csched: delay")
 PERFCOUNTER(vcpu_check,             "csched: vcpu_check")
 PERFCOUNTER(schedule,               "csched: schedule")
 PERFCOUNTER(acct_run,               "csched: acct_run")