6711316: Open source the Garbage-First garbage collector

First mercurial integration of the code for the Garbage-First garbage collector.

Reviewed-by: apetrusenko, iveresov, jmasa, sgoldman, tonyp, ysr

hotspot/src/share/vm/gc_implementation/shared/coTracker.hpp

@@ -0,0 +1,181 @@
+/*
+ * Copyright 2001-2007 Sun Microsystems, Inc.  All Rights Reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
+ *
+ */
+
+// COTracker keeps track of the concurrent overhead of a GC thread.
+
+// A thread that needs to be tracked must, itself, start up its
+// tracker with the start() method and then call the update() method
+// at regular intervals. What the tracker does is to calculate the
+// concurrent overhead of a process at a given update period. The
+// tracker starts and when is detects that it has exceeded the given
+// period, it calculates the duration of the period in wall-clock time
+// and the duration of the period in vtime (i.e. how much time the
+// concurrent processes really took up during this period). The ratio
+// of the latter over the former is the concurrent overhead of that
+// process for that period over a single CPU. This overhead is stored
+// on the tracker, "timestamped" with the wall-clock time of the end
+// of the period. When the concurrent overhead of this process needs
+// to be queried, this last "reading" provides a good approximation
+// (we assume that the concurrent overhead of a particular thread
+// stays largely constant over time). The timestamp is necessary to
+// detect when the process has stopped working and the recorded
+// reading hasn't been updated for some time.
+
+// Each concurrent GC thread is considered to be part of a "group"
+// (i.e. any available concurrent marking threads are part of the
+// "concurrent marking thread group"). A COTracker is associated with
+// a single group at construction-time. It's up to each collector to
+// decide how groups will be mapped to such an id (ids should start
+// from 0 and be consecutive; there's a hardcoded max group num
+// defined on the GCOverheadTracker class). The notion of a group has
+// been introduced to be able to identify how much overhead was
+// imposed by each group, instead of getting a single value that
+// covers all concurrent overhead.
+
+class COTracker {
+private:
+  // It indicates whether this tracker is enabled or not. When the
+  // tracker is disabled, then it returns 0.0 as the latest concurrent
+  // overhead and several methods (reset, start, and update) are not
+  // supposed to be called on it. This enabling / disabling facility
+  // is really provided to make a bit more explicit in the code when a
+  // particulary tracker of a processes that doesn't run all the time
+  // (e.g. concurrent marking) is supposed to be used and not it's not.
+  bool               _enabled;
+
+  // The ID of the group associated with this tracker.
+  int                _group;
+
+  // The update period of the tracker. A new value for the concurrent
+  // overhead of the associated process will be made at intervals no
+  // smaller than this.
+  double             _update_period_sec;
+
+  // The start times (both wall-block time and vtime) of the current
+  // interval.
+  double             _period_start_time_sec;
+  double             _period_start_vtime_sec;
+
+  // Number seq of the concurrent overhead readings within a period
+  NumberSeq          _conc_overhead_seq;
+
+  // The latest reading of the concurrent overhead (over a single CPU)
+  // imposed by the associated concurrent thread, made available at
+  // the indicated wall-clock time.
+  double             _conc_overhead;
+  double             _time_stamp_sec;
+
+  // The number of CPUs that the host machine has (for convenience
+  // really, as we'd have to keep translating it into a double)
+  static double      _cpu_number;
+
+  // Fields that keep a list of all trackers created. This is useful,
+  // since it allows us to sum up the concurrent overhead without
+  // having to write code for a specific collector to broadcast a
+  // request to all its concurrent processes.
+  COTracker*         _next;
+  static COTracker*  _head;
+
+  // It indicates that a new period is starting by updating the
+  // _period_start_time_sec and _period_start_vtime_sec fields.
+  void resetPeriod(double now_sec, double vnow_sec);
+  // It updates the latest concurrent overhead reading, taken at a
+  // given wall-clock time.
+  void setConcOverhead(double time_stamp_sec, double conc_overhead);
+
+  // It determines whether the time stamp of the latest concurrent
+  // overhead reading is out of date or not.
+  bool outOfDate(double now_sec) {
+    // The latest reading is considered out of date, if it was taken
+    // 1.2x the update period.
+    return (now_sec - _time_stamp_sec) > 1.2 * _update_period_sec;
+  }
+
+public:
+  // The constructor which associates the tracker with a group ID.
+  COTracker(int group);
+
+  // Methods to enable / disable the tracker and query whether it is enabled.
+  void enable()  { _enabled = true;  }
+  void disable() { _enabled = false; }
+  bool enabled() { return _enabled;  }
+
+  // It resets the tracker and sets concurrent overhead reading to be
+  // the given parameter and the associated time stamp to be now.
+  void reset(double starting_conc_overhead = 0.0);
+  // The tracker starts tracking. IT should only be called from the
+  // concurrent thread that is tracked by this tracker.
+  void start();
+  // It updates the tracker and, if the current period is longer than
+  // the update period, the concurrent overhead reading will be
+  // updated. force_end being true indicates that it's the last call
+  // to update() by this process before the tracker is disabled (the
+  // tracker can be re-enabled later if necessary).  It should only be
+  // called from the concurrent thread that is tracked by this tracker
+  // and while the thread has joined the STS.
+  void update(bool force_end = false);
+  // It adjusts the contents of the tracker to take into account a STW
+  // pause.
+  void updateForSTW(double start_sec, double end_sec);
+
+  // It returns the last concurrent overhead reading over a single
+  // CPU. If the reading is out of date, or the tracker is disabled,
+  // it returns 0.0.
+  double concCPUOverhead(double now_sec) {
+    if (!_enabled || outOfDate(now_sec))
+      return 0.0;
+    else
+      return _conc_overhead;
+  }
+
+  // It returns the last concurrent overhead reading over all CPUs
+  // that the host machine has. If the reading is out of date, or the
+  // tracker is disabled, it returns 0.0.
+  double concOverhead(double now_sec) {
+    return concCPUOverhead(now_sec) / _cpu_number;
+  }
+
+  double predConcOverhead();
+
+  void resetPred();
+
+  // statics
+
+  // It notifies all trackers about a STW pause.
+  static void updateAllForSTW(double start_sec, double end_sec);
+
+  // It returns the sum of the concurrent overhead readings of all
+  // available (and enabled) trackers for the given time stamp. The
+  // overhead is over all the CPUs of the host machine.
+
+  static double totalConcOverhead(double now_sec);
+  // Like the previous method, but it also sums up the overheads per
+  // group number. The length of the co_per_group array must be at
+  // least as large group_num
+  static double totalConcOverhead(double now_sec,
+                                  size_t group_num,
+                                  double* co_per_group);
+
+  static double totalPredConcOverhead();
+};