1 /*
   2  * Copyright (c) 2016, 2019, Oracle and/or its affiliates. All rights reserved.
   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 Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "gc/g1/g1Analytics.hpp"
  27 #include "gc/g1/g1Predictions.hpp"
  28 #include "runtime/globals.hpp"
  29 #include "runtime/os.hpp"
  30 #include "utilities/debug.hpp"
  31 #include "utilities/globalDefinitions.hpp"
  32 #include "utilities/numberSeq.hpp"
  33 
  34 // Different defaults for different number of GC threads
  35 // They were chosen by running GCOld and SPECjbb on debris with different
  36 //   numbers of GC threads and choosing them based on the results
  37 
  38 // all the same
  39 static double rs_length_diff_defaults[] = {
  40   0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0
  41 };
  42 
  43 static double cost_per_logged_card_ms_defaults[] = {
  44   0.01, 0.005, 0.005, 0.003, 0.003, 0.002, 0.002, 0.0015
  45 };
  46 
  47 // all the same
  48 static double young_card_merge_to_scan_ratio_defaults[] = {
  49   1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0
  50 };
  51 
  52 static double young_only_cost_per_card_scan_ms_defaults[] = {
  53   0.015, 0.01, 0.01, 0.008, 0.008, 0.0055, 0.0055, 0.005
  54 };
  55 
  56 static double cost_per_byte_ms_defaults[] = {
  57   0.00006, 0.00003, 0.00003, 0.000015, 0.000015, 0.00001, 0.00001, 0.000009
  58 };
  59 
  60 // these should be pretty consistent
  61 static double constant_other_time_ms_defaults[] = {
  62   5.0, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0
  63 };
  64 
  65 static double young_other_cost_per_region_ms_defaults[] = {
  66   0.3, 0.2, 0.2, 0.15, 0.15, 0.12, 0.12, 0.1
  67 };
  68 
  69 static double non_young_other_cost_per_region_ms_defaults[] = {
  70   1.0, 0.7, 0.7, 0.5, 0.5, 0.42, 0.42, 0.30
  71 };
  72 
  73 G1Analytics::G1Analytics(const G1Predictions* predictor) :
  74     _predictor(predictor),
  75     _recent_gc_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)),
  76     _concurrent_mark_remark_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)),
  77     _concurrent_mark_cleanup_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)),
  78     _alloc_rate_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
  79     _prev_collection_pause_end_ms(0.0),
  80     _rs_length_diff_seq(new TruncatedSeq(TruncatedSeqLength)),
  81     _concurrent_refine_rate_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
  82     _logged_cards_rate_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
  83     _young_card_merge_to_scan_ratio_seq(new TruncatedSeq(TruncatedSeqLength)),
  84     _mixed_card_merge_to_scan_ratio_seq(new TruncatedSeq(TruncatedSeqLength)),
  85     _young_cost_per_card_scan_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
  86     _mixed_cost_per_card_scan_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
  87     _young_cost_per_card_merge_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
  88     _mixed_cost_per_card_merge_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
  89     _copy_cost_per_byte_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
  90     _constant_other_time_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
  91     _young_other_cost_per_region_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
  92     _non_young_other_cost_per_region_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
  93     _pending_cards_seq(new TruncatedSeq(TruncatedSeqLength)),
  94     _rs_length_seq(new TruncatedSeq(TruncatedSeqLength)),
  95     _cost_per_byte_ms_during_cm_seq(new TruncatedSeq(TruncatedSeqLength)),
  96     _recent_prev_end_times_for_all_gcs_sec(new TruncatedSeq(NumPrevPausesForHeuristics)),
  97     _recent_avg_pause_time_ratio(0.0),
  98     _last_pause_time_ratio(0.0) {
  99 
 100   // Seed sequences with initial values.
 101   _recent_prev_end_times_for_all_gcs_sec->add(os::elapsedTime());
 102   _prev_collection_pause_end_ms = os::elapsedTime() * 1000.0;
 103 
 104   int index = MIN2(ParallelGCThreads - 1, 7u);
 105 
 106   _rs_length_diff_seq->add(rs_length_diff_defaults[index]);
 107   // Start with inverse of maximum STW cost.
 108   _concurrent_refine_rate_ms_seq->add(1/cost_per_logged_card_ms_defaults[0]);
 109   // Some applications have very low rates for logging cards.
 110   _logged_cards_rate_ms_seq->add(0.0);
 111   _young_card_merge_to_scan_ratio_seq->add(young_card_merge_to_scan_ratio_defaults[index]);
 112   _young_cost_per_card_scan_ms_seq->add(young_only_cost_per_card_scan_ms_defaults[index]);
 113 
 114   _copy_cost_per_byte_ms_seq->add(cost_per_byte_ms_defaults[index]);
 115   _constant_other_time_ms_seq->add(constant_other_time_ms_defaults[index]);
 116   _young_other_cost_per_region_ms_seq->add(young_other_cost_per_region_ms_defaults[index]);
 117   _non_young_other_cost_per_region_ms_seq->add(non_young_other_cost_per_region_ms_defaults[index]);
 118 
 119   // start conservatively (around 50ms is about right)
 120   _concurrent_mark_remark_times_ms->add(0.05);
 121   _concurrent_mark_cleanup_times_ms->add(0.20);
 122 }
 123 
 124 bool G1Analytics::enough_samples_available(TruncatedSeq const* seq) const {
 125   return seq->num() >= 3;
 126 }
 127 
 128 double G1Analytics::get_new_prediction(TruncatedSeq const* seq) const {
 129   return _predictor->get_new_prediction(seq);
 130 }
 131 
 132 size_t G1Analytics::get_new_size_prediction(TruncatedSeq const* seq) const {
 133   return (size_t)get_new_prediction(seq);
 134 }
 135 
 136 int G1Analytics::num_alloc_rate_ms() const {
 137   return _alloc_rate_ms_seq->num();
 138 }
 139 
 140 void G1Analytics::report_concurrent_mark_remark_times_ms(double ms) {
 141   _concurrent_mark_remark_times_ms->add(ms);
 142 }
 143 
 144 void G1Analytics::report_alloc_rate_ms(double alloc_rate) {
 145   _alloc_rate_ms_seq->add(alloc_rate);
 146 }
 147 
 148 void G1Analytics::compute_pause_time_ratio(double interval_ms, double pause_time_ms) {
 149   _recent_avg_pause_time_ratio = _recent_gc_times_ms->sum() / interval_ms;
 150 
 151   // Clamp the result to [0.0 ... 1.0] to filter out nonsensical results due to bad input.
 152   _recent_avg_pause_time_ratio = clamp(_recent_avg_pause_time_ratio, 0.0, 1.0);
 153 
 154   // Compute the ratio of just this last pause time to the entire time range stored
 155   // in the vectors. Comparing this pause to the entire range, rather than only the
 156   // most recent interval, has the effect of smoothing over a possible transient 'burst'
 157   // of more frequent pauses that don't really reflect a change in heap occupancy.
 158   // This reduces the likelihood of a needless heap expansion being triggered.
 159   _last_pause_time_ratio =
 160     (pause_time_ms * _recent_prev_end_times_for_all_gcs_sec->num()) / interval_ms;
 161 }
 162 
 163 void G1Analytics::report_concurrent_refine_rate_ms(double cards_per_ms) {
 164   _concurrent_refine_rate_ms_seq->add(cards_per_ms);
 165 }
 166 
 167 void G1Analytics::report_logged_cards_rate_ms(double cards_per_ms) {
 168   _logged_cards_rate_ms_seq->add(cards_per_ms);
 169 }
 170 
 171 void G1Analytics::report_cost_per_card_scan_ms(double cost_per_card_ms, bool for_young_gc) {
 172   if (for_young_gc) {
 173     _young_cost_per_card_scan_ms_seq->add(cost_per_card_ms);
 174   } else {
 175     _mixed_cost_per_card_scan_ms_seq->add(cost_per_card_ms);
 176   }
 177 }
 178 
 179 void G1Analytics::report_cost_per_card_merge_ms(double cost_per_card_ms, bool for_young_gc) {
 180   if (for_young_gc) {
 181     _young_cost_per_card_merge_ms_seq->add(cost_per_card_ms);
 182   } else {
 183     _mixed_cost_per_card_merge_ms_seq->add(cost_per_card_ms);
 184   }
 185 }
 186 
 187 void G1Analytics::report_card_merge_to_scan_ratio(double merge_to_scan_ratio, bool for_young_gc) {
 188   if (for_young_gc) {
 189     _young_card_merge_to_scan_ratio_seq->add(merge_to_scan_ratio);
 190   } else {
 191     _mixed_card_merge_to_scan_ratio_seq->add(merge_to_scan_ratio);
 192   }
 193 }
 194 
 195 void G1Analytics::report_rs_length_diff(double rs_length_diff) {
 196   _rs_length_diff_seq->add(rs_length_diff);
 197 }
 198 
 199 void G1Analytics::report_cost_per_byte_ms(double cost_per_byte_ms, bool mark_or_rebuild_in_progress) {
 200   if (mark_or_rebuild_in_progress) {
 201     _cost_per_byte_ms_during_cm_seq->add(cost_per_byte_ms);
 202   } else {
 203     _copy_cost_per_byte_ms_seq->add(cost_per_byte_ms);
 204   }
 205 }
 206 
 207 void G1Analytics::report_young_other_cost_per_region_ms(double other_cost_per_region_ms) {
 208   _young_other_cost_per_region_ms_seq->add(other_cost_per_region_ms);
 209 }
 210 
 211 void G1Analytics::report_non_young_other_cost_per_region_ms(double other_cost_per_region_ms) {
 212   _non_young_other_cost_per_region_ms_seq->add(other_cost_per_region_ms);
 213 }
 214 
 215 void G1Analytics::report_constant_other_time_ms(double constant_other_time_ms) {
 216   _constant_other_time_ms_seq->add(constant_other_time_ms);
 217 }
 218 
 219 void G1Analytics::report_pending_cards(double pending_cards) {
 220   _pending_cards_seq->add(pending_cards);
 221 }
 222 
 223 void G1Analytics::report_rs_length(double rs_length) {
 224   _rs_length_seq->add(rs_length);
 225 }
 226 
 227 double G1Analytics::predict_alloc_rate_ms() const {
 228   return get_new_prediction(_alloc_rate_ms_seq);
 229 }
 230 
 231 double G1Analytics::predict_concurrent_refine_rate_ms() const {
 232   return get_new_prediction(_concurrent_refine_rate_ms_seq);
 233 }
 234 
 235 double G1Analytics::predict_logged_cards_rate_ms() const {
 236   return get_new_prediction(_logged_cards_rate_ms_seq);
 237 }
 238 
 239 double G1Analytics::predict_young_card_merge_to_scan_ratio() const {
 240   return get_new_prediction(_young_card_merge_to_scan_ratio_seq);
 241 }
 242 
 243 size_t G1Analytics::predict_scan_card_num(size_t rs_length, bool for_young_gc) const {
 244   if (for_young_gc || !enough_samples_available(_mixed_card_merge_to_scan_ratio_seq)) {
 245     return (size_t) (rs_length * predict_young_card_merge_to_scan_ratio());
 246   } else {
 247     return (size_t) (rs_length * get_new_prediction(_mixed_card_merge_to_scan_ratio_seq));
 248   }
 249 }
 250 
 251 double G1Analytics::predict_card_merge_time_ms(size_t card_num, bool for_young_gc) const {
 252   if (for_young_gc || !enough_samples_available(_mixed_cost_per_card_merge_ms_seq)) {
 253     return card_num * get_new_prediction(_young_cost_per_card_merge_ms_seq);
 254   } else {
 255     return card_num * get_new_prediction(_mixed_cost_per_card_merge_ms_seq);
 256   }
 257 }
 258 
 259 double G1Analytics::predict_card_scan_time_ms(size_t card_num, bool for_young_gc) const {
 260   if (for_young_gc || !enough_samples_available(_mixed_cost_per_card_scan_ms_seq)) {
 261     return card_num * get_new_prediction(_young_cost_per_card_scan_ms_seq);
 262   } else {
 263     return card_num * get_new_prediction(_mixed_cost_per_card_scan_ms_seq);
 264   }
 265 }
 266 
 267 double G1Analytics::predict_object_copy_time_ms_during_cm(size_t bytes_to_copy) const {
 268   if (!enough_samples_available(_cost_per_byte_ms_during_cm_seq)) {
 269     return (1.1 * bytes_to_copy) * get_new_prediction(_copy_cost_per_byte_ms_seq);
 270   } else {
 271     return bytes_to_copy * get_new_prediction(_cost_per_byte_ms_during_cm_seq);
 272   }
 273 }
 274 
 275 double G1Analytics::predict_object_copy_time_ms(size_t bytes_to_copy, bool during_concurrent_mark) const {
 276   if (during_concurrent_mark) {
 277     return predict_object_copy_time_ms_during_cm(bytes_to_copy);
 278   } else {
 279     return bytes_to_copy * get_new_prediction(_copy_cost_per_byte_ms_seq);
 280   }
 281 }
 282 
 283 double G1Analytics::predict_constant_other_time_ms() const {
 284   return get_new_prediction(_constant_other_time_ms_seq);
 285 }
 286 
 287 double G1Analytics::predict_young_other_time_ms(size_t young_num) const {
 288   return young_num * get_new_prediction(_young_other_cost_per_region_ms_seq);
 289 }
 290 
 291 double G1Analytics::predict_non_young_other_time_ms(size_t non_young_num) const {
 292   return non_young_num * get_new_prediction(_non_young_other_cost_per_region_ms_seq);
 293 }
 294 
 295 double G1Analytics::predict_remark_time_ms() const {
 296   return get_new_prediction(_concurrent_mark_remark_times_ms);
 297 }
 298 
 299 double G1Analytics::predict_cleanup_time_ms() const {
 300   return get_new_prediction(_concurrent_mark_cleanup_times_ms);
 301 }
 302 
 303 size_t G1Analytics::predict_rs_length() const {
 304   return get_new_size_prediction(_rs_length_seq) + get_new_prediction(_rs_length_diff_seq);
 305 }
 306 
 307 size_t G1Analytics::predict_pending_cards() const {
 308   return get_new_size_prediction(_pending_cards_seq);
 309 }
 310 
 311 double G1Analytics::last_known_gc_end_time_sec() const {
 312   return _recent_prev_end_times_for_all_gcs_sec->oldest();
 313 }
 314 
 315 void G1Analytics::update_recent_gc_times(double end_time_sec,
 316                                          double pause_time_ms) {
 317   _recent_gc_times_ms->add(pause_time_ms);
 318   _recent_prev_end_times_for_all_gcs_sec->add(end_time_sec);
 319   _prev_collection_pause_end_ms = end_time_sec * 1000.0;
 320 }
 321 
 322 void G1Analytics::report_concurrent_mark_cleanup_times_ms(double ms) {
 323   _concurrent_mark_cleanup_times_ms->add(ms);
 324 }