i#6938 sched migrate: Separate run queue per output

clients/drcachesim/analyzer_multi.cpp

@@ -553,13 +553,20 @@ analyzer_multi_tmpl_t<RecordType, ReaderType>::init_dynamic_schedule()
         op_sched_order_time.get_value() ? sched_type_t::DEPENDENCY_TIMESTAMPS
                                         : sched_type_t::DEPENDENCY_IGNORE,
         sched_type_t::SCHEDULER_DEFAULTS, op_verbose.get_value());
-    sched_ops.quantum_duration = op_sched_quantum.get_value();
-    if (op_sched_time.get_value())
+    sched_ops.time_units_per_us = op_sched_time_per_us.get_value();
+    if (op_sched_time.get_value()) {
         sched_ops.quantum_unit = sched_type_t::QUANTUM_TIME;
+        sched_ops.quantum_duration_us = op_sched_quantum.get_value();
+    } else {
+        sched_ops.quantum_duration_instrs = op_sched_quantum.get_value();
+    }
     sched_ops.syscall_switch_threshold = op_sched_syscall_switch_us.get_value();
     sched_ops.blocking_switch_threshold = op_sched_blocking_switch_us.get_value();
-    sched_ops.block_time_scale = op_sched_block_scale.get_value();
-    sched_ops.block_time_max = op_sched_block_max_us.get_value();
+    sched_ops.block_time_multiplier = op_sched_block_scale.get_value();
+    sched_ops.block_time_max_us = op_sched_block_max_us.get_value();
+    sched_ops.migration_threshold_us = op_sched_migration_threshold_us.get_value();
+    sched_ops.rebalance_period_us = op_sched_rebalance_period_us.get_value();
+    sched_ops.time_units_per_us = op_sched_time_units_per_us.get_value();
     sched_ops.randomize_next_input = op_sched_randomize.get_value();
     sched_ops.honor_direct_switches = !op_sched_disable_direct_switches.get_value();
 #ifdef HAS_ZIP

clients/drcachesim/common/memtrace_stream.h

@@ -98,6 +98,15 @@ class memtrace_stream_t {
          * i.e., the number of input migrations to this core.
          */
         SCHED_STAT_MIGRATIONS,
+        /**
+         * Counts the number of times this output's runqueue became empty and it took
+         * work from another output's runqueue.
+         */
+        SCHED_STAT_RUNQUEUE_STEALS,
+        /**
+         * Counts the number of output runqueue rebalances triggered by this output.
+         */
+        SCHED_STAT_RUNQUEUE_REBALANCES,
         /** Count of statistic types. */
         SCHED_STAT_TYPE_COUNT,
     };

clients/drcachesim/common/options.cpp

@@ -887,13 +887,19 @@ droption_t<bool> op_core_serial(
     "How the scheduling is performed is controlled by a set "
     "of options with the prefix \"sched_\" along with -cores.");
 
+droption_t<double>
+    op_sched_time_per_us(DROPTION_SCOPE_ALL, "sched_time_per_us", 1000.,
+                         "Wall-clock microseconds per simulated microsecond",
+                         "Wall-clock microseconds per simulated microsecond.");
+
 droption_t<int64_t>
     // We pick 6 million to match 2 instructions per nanosecond with a 3ms quantum.
     op_sched_quantum(DROPTION_SCOPE_ALL, "sched_quantum", 6 * 1000 * 1000,
                      "Scheduling quantum",
                      "Applies to -core_sharded and -core_serial. "
-                     "Scheduling quantum: in microseconds of wall-clock "
-                     "time if -sched_time is set; otherwise in instructions.");
+                     "Scheduling quantum in instructions, unless -sched_time is set in "
+                     "which case this value is multiplied by -sched_time_per_us to "
+                     "produce a quantum in wall-clock microseconds.");
 
 droption_t<bool>
     op_sched_time(DROPTION_SCOPE_ALL, "sched_time", false,
@@ -922,23 +928,24 @@ droption_t<uint64_t> op_sched_blocking_switch_us(
     "-core_serial. ");
 
 droption_t<double> op_sched_block_scale(
-    DROPTION_SCOPE_ALL, "sched_block_scale", 10., "Input block time scale factor",
-    "The scale applied to the microsecond latency of blocking system calls.  A higher "
-    "value here results in blocking syscalls keeping inputs unscheduled for longer.  "
-    "This should roughly equal the slowdown of instruction record processing versus the "
-    "original (untraced) application execution.");
-
-// We have a max to avoid outlier latencies that are already a second or more from
-// scaling up to tens of minutes.  We assume a cap is representative as the outliers
+    DROPTION_SCOPE_ALL, "sched_block_scale", 0.01, "Input block time scale factor",
+    "This value is multiplied by -sched_time_per_us to produce a scale which is applied "
+    "to the as-traced microsecond latency of blocking system calls to produce the block "
+    "time during simulation.  A higher value here results in blocking syscalls keeping "
+    "inputs unscheduled for longer.");
+
+// We have a max to avoid outlier latencies from scaling up to extreme times.  There is
+// some inflation in the as-traced latencies and some can be inflated more than others.
+// We assume a cap is representative as the outliers
 // likely were not part of key dependence chains.  Without a cap the other threads all
 // finish and the simulation waits for tens of minutes further for a couple of outliers.
 // The cap remains a flag and not a constant as different length traces and different
 // speed simulators need different idle time ranges, so we need to be able to tune this
 // to achieve desired cpu usage targets.  The default value was selected to avoid unduly
 // long idle times with local analyzers; it may need to be increased with more
 // heavyweight analyzers/simulators.
-droption_t<uint64_t> op_sched_block_max_us(DROPTION_SCOPE_ALL, "sched_block_max_us",
-                                           250000,
+// TODO i#6959: Once we have -exit_if_all_unscheduled raise this.
+droption_t<uint64_t> op_sched_block_max_us(DROPTION_SCOPE_ALL, "sched_block_max_us", 250,
                                            "Maximum blocked input time, in microseconds",
                                            "The maximum blocked time, after scaling with "
                                            "-sched_block_scale.");
@@ -985,6 +992,25 @@ droption_t<bool> op_sched_disable_direct_switches(
     "switch being determined by latency and the next input in the queue.  The "
     "TRACE_MARKER_TYPE_DIRECT_THREAD_SWITCH markers are not removed from the trace.");
 
+droption_t<uint64_t> op_sched_migration_threshold_us(
+    DROPTION_SCOPE_ALL, "sched_migration_threshold_us", 500,
+    "Time in simulated microseconds before an input can be migrated across cores",
+    "The minimum time in simulated microseconds that must have elapsed since an input "
+    "last ran on a core before it can be migrated to another core.");
+
+droption_t<uint64_t> op_sched_rebalance_period_us(
+    DROPTION_SCOPE_ALL, "sched_rebalance_period_us", 1500000,
+    "Period in microseconds at which core run queues are load-balanced",
+    "The period in simulated microseconds at which per-core run queues are re-balanced "
+    "to redistribute load.");
+
+droption_t<double> op_sched_time_units_per_us(
+    DROPTION_SCOPE_ALL, "sched_time_units_per_us", 1000.,
+    "Time units per simulated microsecond",
+    "Time units (currently wall-clock time) per simulated microsecond.  This scales all "
+    "of the -sched_*_us values as it concerts wall-clock time into the simulated "
+    "microseconds measured by those options.");
+
 // Schedule_stats options.
 droption_t<uint64_t>
     op_schedule_stats_print_every(DROPTION_SCOPE_ALL, "schedule_stats_print_every",

clients/drcachesim/common/options.h

@@ -199,6 +199,7 @@ extern dynamorio::droption::droption_t<int> op_kernel_trace_buffer_size_shift;
 #endif
 extern dynamorio::droption::droption_t<bool> op_core_sharded;
 extern dynamorio::droption::droption_t<bool> op_core_serial;
+extern dynamorio::droption::droption_t<double> op_sched_time_per_us;
 extern dynamorio::droption::droption_t<int64_t> op_sched_quantum;
 extern dynamorio::droption::droption_t<bool> op_sched_time;
 extern dynamorio::droption::droption_t<bool> op_sched_order_time;
@@ -214,6 +215,9 @@ extern dynamorio::droption::droption_t<std::string> op_cpu_schedule_file;
 extern dynamorio::droption::droption_t<std::string> op_sched_switch_file;
 extern dynamorio::droption::droption_t<bool> op_sched_randomize;
 extern dynamorio::droption::droption_t<bool> op_sched_disable_direct_switches;
+extern dynamorio::droption::droption_t<uint64_t> op_sched_migration_threshold_us;
+extern dynamorio::droption::droption_t<uint64_t> op_sched_rebalance_period_us;
+extern dynamorio::droption::droption_t<double> op_sched_time_units_per_us;
 extern dynamorio::droption::droption_t<uint64_t> op_schedule_stats_print_every;
 extern dynamorio::droption::droption_t<std::string> op_syscall_template_file;
 extern dynamorio::droption::droption_t<uint64_t> op_filter_stop_timestamp;

clients/drcachesim/scheduler/flexible_queue.h

@@ -108,6 +108,15 @@ class flexible_queue_t {
         return entries_[rand_gen_() % size()]; // Undefined if empty.
     }
 
+    // Returns an entry from the back -- or at least not from the front; it's not
+    // guaranteed to be the lowest priority, just not the highest.
+    T
+    back()
+    {
+        assert(!empty());
+        return entries_.back();
+    }
+
     bool
     empty() const
     {