Mercurial > hg > truffle
annotate src/share/vm/gc_implementation/shared/gcUtil.hpp @ 1145:e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
Summary: Autonomic per-worker free block cache sizing, tunable coalition policies, fixes to per-size block statistics, retuned gain and bandwidth of some feedback loop filters to allow quicker reactivity to abrupt changes in ambient demand, and other heuristics to reduce fragmentation of the CMS old gen. Also tightened some assertions, including those related to locking.
Reviewed-by: jmasa
| author | ysr |
|---|---|
| date | Wed, 23 Dec 2009 09:23:54 -0800 |
| parents | 9ee9cf798b59 |
| children | c18cbe5936b8 |
| rev | line source |
|---|---|
| 0 | 1 /* |
| 337 | 2 * Copyright 2002-2008 Sun Microsystems, Inc. All Rights Reserved. |
| 0 | 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
| 4 * | |
| 5 * This code is free software; you can redistribute it and/or modify it | |
| 6 * under the terms of the GNU General Public License version 2 only, as | |
| 7 * published by the Free Software Foundation. | |
| 8 * | |
| 9 * This code is distributed in the hope that it will be useful, but WITHOUT | |
| 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
| 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
| 12 * version 2 for more details (a copy is included in the LICENSE file that | |
| 13 * accompanied this code). | |
| 14 * | |
| 15 * You should have received a copy of the GNU General Public License version | |
| 16 * 2 along with this work; if not, write to the Free Software Foundation, | |
| 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. | |
| 18 * | |
| 19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
| 20 * CA 95054 USA or visit www.sun.com if you need additional information or | |
| 21 * have any questions. | |
| 22 * | |
| 23 */ | |
| 24 | |
| 25 // Catch-all file for utility classes | |
| 26 | |
| 27 // A weighted average maintains a running, weighted average | |
| 28 // of some float value (templates would be handy here if we | |
| 29 // need different types). | |
| 30 // | |
| 31 // The average is adaptive in that we smooth it for the | |
| 32 // initial samples; we don't use the weight until we have | |
| 33 // enough samples for it to be meaningful. | |
| 34 // | |
| 35 // This serves as our best estimate of a future unknown. | |
| 36 // | |
| 37 class AdaptiveWeightedAverage : public CHeapObj { | |
| 38 private: | |
| 39 float _average; // The last computed average | |
| 40 unsigned _sample_count; // How often we've sampled this average | |
| 41 unsigned _weight; // The weight used to smooth the averages | |
| 42 // A higher weight favors the most | |
| 43 // recent data. | |
| 44 | |
| 45 protected: | |
| 46 float _last_sample; // The last value sampled. | |
| 47 | |
| 48 void increment_count() { _sample_count++; } | |
| 49 void set_average(float avg) { _average = avg; } | |
| 50 | |
| 51 // Helper function, computes an adaptive weighted average | |
| 52 // given a sample and the last average | |
| 53 float compute_adaptive_average(float new_sample, float average); | |
| 54 | |
| 55 public: | |
| 56 // Input weight must be between 0 and 100 | |
|
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
57 AdaptiveWeightedAverage(unsigned weight, float avg = 0.0) : |
|
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
58 _average(avg), _sample_count(0), _weight(weight), _last_sample(0.0) { |
| 0 | 59 } |
| 60 | |
|
268
d6340ab4105b
6723228: NUMA allocator: assert(lgrp_id != -1, "No lgrp_id set")
iveresov
parents:
0
diff
changeset
|
61 void clear() { |
|
d6340ab4105b
6723228: NUMA allocator: assert(lgrp_id != -1, "No lgrp_id set")
iveresov
parents:
0
diff
changeset
|
62 _average = 0; |
|
d6340ab4105b
6723228: NUMA allocator: assert(lgrp_id != -1, "No lgrp_id set")
iveresov
parents:
0
diff
changeset
|
63 _sample_count = 0; |
|
d6340ab4105b
6723228: NUMA allocator: assert(lgrp_id != -1, "No lgrp_id set")
iveresov
parents:
0
diff
changeset
|
64 _last_sample = 0; |
|
d6340ab4105b
6723228: NUMA allocator: assert(lgrp_id != -1, "No lgrp_id set")
iveresov
parents:
0
diff
changeset
|
65 } |
|
d6340ab4105b
6723228: NUMA allocator: assert(lgrp_id != -1, "No lgrp_id set")
iveresov
parents:
0
diff
changeset
|
66 |
|
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
67 // Useful for modifying static structures after startup. |
|
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
68 void modify(size_t avg, unsigned wt, bool force = false) { |
|
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
69 assert(force, "Are you sure you want to call this?"); |
|
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
70 _average = (float)avg; |
|
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
71 _weight = wt; |
|
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
72 } |
|
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
73 |
| 0 | 74 // Accessors |
| 75 float average() const { return _average; } | |
| 76 unsigned weight() const { return _weight; } | |
| 77 unsigned count() const { return _sample_count; } | |
| 78 float last_sample() const { return _last_sample; } | |
| 79 | |
| 80 // Update data with a new sample. | |
| 81 void sample(float new_sample); | |
| 82 | |
| 83 static inline float exp_avg(float avg, float sample, | |
| 84 unsigned int weight) { | |
| 85 assert(0 <= weight && weight <= 100, "weight must be a percent"); | |
| 86 return (100.0F - weight) * avg / 100.0F + weight * sample / 100.0F; | |
| 87 } | |
| 88 static inline size_t exp_avg(size_t avg, size_t sample, | |
| 89 unsigned int weight) { | |
| 90 // Convert to float and back to avoid integer overflow. | |
| 91 return (size_t)exp_avg((float)avg, (float)sample, weight); | |
| 92 } | |
|
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
93 |
|
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
94 // Printing |
|
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
95 void print_on(outputStream* st) const; |
|
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
96 void print() const; |
| 0 | 97 }; |
| 98 | |
| 99 | |
| 100 // A weighted average that includes a deviation from the average, | |
| 101 // some multiple of which is added to the average. | |
| 102 // | |
| 103 // This serves as our best estimate of an upper bound on a future | |
| 104 // unknown. | |
| 105 class AdaptivePaddedAverage : public AdaptiveWeightedAverage { | |
| 106 private: | |
| 107 float _padded_avg; // The last computed padded average | |
| 108 float _deviation; // Running deviation from the average | |
| 109 unsigned _padding; // A multiple which, added to the average, | |
| 110 // gives us an upper bound guess. | |
| 111 | |
| 112 protected: | |
| 113 void set_padded_average(float avg) { _padded_avg = avg; } | |
| 114 void set_deviation(float dev) { _deviation = dev; } | |
| 115 | |
| 116 public: | |
| 117 AdaptivePaddedAverage() : | |
| 118 AdaptiveWeightedAverage(0), | |
| 119 _padded_avg(0.0), _deviation(0.0), _padding(0) {} | |
| 120 | |
| 121 AdaptivePaddedAverage(unsigned weight, unsigned padding) : | |
| 122 AdaptiveWeightedAverage(weight), | |
| 123 _padded_avg(0.0), _deviation(0.0), _padding(padding) {} | |
| 124 | |
| 125 // Placement support | |
| 126 void* operator new(size_t ignored, void* p) { return p; } | |
| 127 // Allocator | |
| 128 void* operator new(size_t size) { return CHeapObj::operator new(size); } | |
| 129 | |
| 130 // Accessor | |
| 131 float padded_average() const { return _padded_avg; } | |
| 132 float deviation() const { return _deviation; } | |
| 133 unsigned padding() const { return _padding; } | |
| 134 | |
|
268
d6340ab4105b
6723228: NUMA allocator: assert(lgrp_id != -1, "No lgrp_id set")
iveresov
parents:
0
diff
changeset
|
135 void clear() { |
|
d6340ab4105b
6723228: NUMA allocator: assert(lgrp_id != -1, "No lgrp_id set")
iveresov
parents:
0
diff
changeset
|
136 AdaptiveWeightedAverage::clear(); |
|
d6340ab4105b
6723228: NUMA allocator: assert(lgrp_id != -1, "No lgrp_id set")
iveresov
parents:
0
diff
changeset
|
137 _padded_avg = 0; |
|
d6340ab4105b
6723228: NUMA allocator: assert(lgrp_id != -1, "No lgrp_id set")
iveresov
parents:
0
diff
changeset
|
138 _deviation = 0; |
|
d6340ab4105b
6723228: NUMA allocator: assert(lgrp_id != -1, "No lgrp_id set")
iveresov
parents:
0
diff
changeset
|
139 } |
|
d6340ab4105b
6723228: NUMA allocator: assert(lgrp_id != -1, "No lgrp_id set")
iveresov
parents:
0
diff
changeset
|
140 |
| 0 | 141 // Override |
| 142 void sample(float new_sample); | |
|
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
143 |
|
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
144 // Printing |
|
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
145 void print_on(outputStream* st) const; |
|
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
146 void print() const; |
| 0 | 147 }; |
| 148 | |
| 149 // A weighted average that includes a deviation from the average, | |
| 150 // some multiple of which is added to the average. | |
| 151 // | |
| 152 // This serves as our best estimate of an upper bound on a future | |
| 153 // unknown. | |
| 154 // A special sort of padded average: it doesn't update deviations | |
| 155 // if the sample is zero. The average is allowed to change. We're | |
| 156 // preventing the zero samples from drastically changing our padded | |
| 157 // average. | |
| 158 class AdaptivePaddedNoZeroDevAverage : public AdaptivePaddedAverage { | |
| 159 public: | |
| 160 AdaptivePaddedNoZeroDevAverage(unsigned weight, unsigned padding) : | |
| 161 AdaptivePaddedAverage(weight, padding) {} | |
| 162 // Override | |
| 163 void sample(float new_sample); | |
|
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
164 |
|
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
165 // Printing |
|
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
166 void print_on(outputStream* st) const; |
|
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
167 void print() const; |
| 0 | 168 }; |
|
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
337
diff
changeset
|
169 |
| 0 | 170 // Use a least squares fit to a set of data to generate a linear |
| 171 // equation. | |
| 172 // y = intercept + slope * x | |
| 173 | |
| 174 class LinearLeastSquareFit : public CHeapObj { | |
| 175 double _sum_x; // sum of all independent data points x | |
| 176 double _sum_x_squared; // sum of all independent data points x**2 | |
| 177 double _sum_y; // sum of all dependent data points y | |
| 178 double _sum_xy; // sum of all x * y. | |
| 179 double _intercept; // constant term | |
| 180 double _slope; // slope | |
| 181 // The weighted averages are not currently used but perhaps should | |
| 182 // be used to get decaying averages. | |
| 183 AdaptiveWeightedAverage _mean_x; // weighted mean of independent variable | |
| 184 AdaptiveWeightedAverage _mean_y; // weighted mean of dependent variable | |
| 185 | |
| 186 public: | |
| 187 LinearLeastSquareFit(unsigned weight); | |
| 188 void update(double x, double y); | |
| 189 double y(double x); | |
| 190 double slope() { return _slope; } | |
| 191 // Methods to decide if a change in the dependent variable will | |
| 192 // achive a desired goal. Note that these methods are not | |
| 193 // complementary and both are needed. | |
| 194 bool decrement_will_decrease(); | |
| 195 bool increment_will_decrease(); | |
| 196 }; | |
| 197 | |
| 198 class GCPauseTimer : StackObj { | |
| 199 elapsedTimer* _timer; | |
| 200 public: | |
| 201 GCPauseTimer(elapsedTimer* timer) { | |
| 202 _timer = timer; | |
| 203 _timer->stop(); | |
| 204 } | |
| 205 ~GCPauseTimer() { | |
| 206 _timer->start(); | |
| 207 } | |
| 208 }; |