hotspot/src/share/vm/gc_implementation/concurrentMarkSweep/cmsAdaptiveSizePolicy.cpp
7131629: Generalize the CMS free list code
Summary: Make the FreeChunk, FreeList, TreeList, and BinaryTreeDictionary classes usable outside CMS.
Reviewed-by: brutisso, johnc, jwilhelm
Contributed-by: coleen.phillimore@oracle.com
/*
* Copyright (c) 2004, 2010, Oracle and/or its affiliates. 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 Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
* or visit www.oracle.com if you need additional information or have any
* questions.
*
*/
#include "precompiled.hpp"
#include "gc_implementation/concurrentMarkSweep/cmsAdaptiveSizePolicy.hpp"
#include "gc_implementation/shared/gcStats.hpp"
#include "memory/defNewGeneration.hpp"
#include "memory/genCollectedHeap.hpp"
#include "runtime/thread.hpp"
#ifdef TARGET_OS_FAMILY_linux
# include "os_linux.inline.hpp"
#endif
#ifdef TARGET_OS_FAMILY_solaris
# include "os_solaris.inline.hpp"
#endif
#ifdef TARGET_OS_FAMILY_windows
# include "os_windows.inline.hpp"
#endif
#ifdef TARGET_OS_FAMILY_bsd
# include "os_bsd.inline.hpp"
#endif
elapsedTimer CMSAdaptiveSizePolicy::_concurrent_timer;
elapsedTimer CMSAdaptiveSizePolicy::_STW_timer;
// Defined if the granularity of the time measurements is potentially too large.
#define CLOCK_GRANULARITY_TOO_LARGE
CMSAdaptiveSizePolicy::CMSAdaptiveSizePolicy(size_t init_eden_size,
size_t init_promo_size,
size_t init_survivor_size,
double max_gc_minor_pause_sec,
double max_gc_pause_sec,
uint gc_cost_ratio) :
AdaptiveSizePolicy(init_eden_size,
init_promo_size,
init_survivor_size,
max_gc_pause_sec,
gc_cost_ratio) {
clear_internal_time_intervals();
_processor_count = os::active_processor_count();
if (CMSConcurrentMTEnabled && (ConcGCThreads > 1)) {
assert(_processor_count > 0, "Processor count is suspect");
_concurrent_processor_count = MIN2((uint) ConcGCThreads,
(uint) _processor_count);
} else {
_concurrent_processor_count = 1;
}
_avg_concurrent_time = new AdaptiveWeightedAverage(AdaptiveTimeWeight);
_avg_concurrent_interval = new AdaptiveWeightedAverage(AdaptiveTimeWeight);
_avg_concurrent_gc_cost = new AdaptiveWeightedAverage(AdaptiveTimeWeight);
_avg_initial_pause = new AdaptivePaddedAverage(AdaptiveTimeWeight,
PausePadding);
_avg_remark_pause = new AdaptivePaddedAverage(AdaptiveTimeWeight,
PausePadding);
_avg_cms_STW_time = new AdaptiveWeightedAverage(AdaptiveTimeWeight);
_avg_cms_STW_gc_cost = new AdaptiveWeightedAverage(AdaptiveTimeWeight);
_avg_cms_free = new AdaptiveWeightedAverage(AdaptiveTimeWeight);
_avg_cms_free_at_sweep = new AdaptiveWeightedAverage(AdaptiveTimeWeight);
_avg_cms_promo = new AdaptiveWeightedAverage(AdaptiveTimeWeight);
// Mark-sweep-compact
_avg_msc_pause = new AdaptiveWeightedAverage(AdaptiveTimeWeight);
_avg_msc_interval = new AdaptiveWeightedAverage(AdaptiveTimeWeight);
_avg_msc_gc_cost = new AdaptiveWeightedAverage(AdaptiveTimeWeight);
// Mark-sweep
_avg_ms_pause = new AdaptiveWeightedAverage(AdaptiveTimeWeight);
_avg_ms_interval = new AdaptiveWeightedAverage(AdaptiveTimeWeight);
_avg_ms_gc_cost = new AdaptiveWeightedAverage(AdaptiveTimeWeight);
// Variables that estimate pause times as a function of generation
// size.
_remark_pause_old_estimator =
new LinearLeastSquareFit(AdaptiveSizePolicyWeight);
_initial_pause_old_estimator =
new LinearLeastSquareFit(AdaptiveSizePolicyWeight);
_remark_pause_young_estimator =
new LinearLeastSquareFit(AdaptiveSizePolicyWeight);
_initial_pause_young_estimator =
new LinearLeastSquareFit(AdaptiveSizePolicyWeight);
// Alignment comes from that used in ReservedSpace.
_generation_alignment = os::vm_allocation_granularity();
// Start the concurrent timer here so that the first
// concurrent_phases_begin() measures a finite mutator
// time. A finite mutator time is used to determine
// if a concurrent collection has been started. If this
// proves to be a problem, use some explicit flag to
// signal that a concurrent collection has been started.
_concurrent_timer.start();
_STW_timer.start();
}
double CMSAdaptiveSizePolicy::concurrent_processor_fraction() {
// For now assume no other daemon threads are taking alway
// cpu's from the application.
return ((double) _concurrent_processor_count / (double) _processor_count);
}
double CMSAdaptiveSizePolicy::concurrent_collection_cost(
double interval_in_seconds) {
// When the precleaning and sweeping phases use multiple
// threads, change one_processor_fraction to
// concurrent_processor_fraction().
double one_processor_fraction = 1.0 / ((double) processor_count());
double concurrent_cost =
collection_cost(_latest_cms_concurrent_marking_time_secs,
interval_in_seconds) * concurrent_processor_fraction() +
collection_cost(_latest_cms_concurrent_precleaning_time_secs,
interval_in_seconds) * one_processor_fraction +
collection_cost(_latest_cms_concurrent_sweeping_time_secs,
interval_in_seconds) * one_processor_fraction;
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print_cr(
"\nCMSAdaptiveSizePolicy::scaled_concurrent_collection_cost(%f) "
"_latest_cms_concurrent_marking_cost %f "
"_latest_cms_concurrent_precleaning_cost %f "
"_latest_cms_concurrent_sweeping_cost %f "
"concurrent_processor_fraction %f "
"concurrent_cost %f ",
interval_in_seconds,
collection_cost(_latest_cms_concurrent_marking_time_secs,
interval_in_seconds),
collection_cost(_latest_cms_concurrent_precleaning_time_secs,
interval_in_seconds),
collection_cost(_latest_cms_concurrent_sweeping_time_secs,
interval_in_seconds),
concurrent_processor_fraction(),
concurrent_cost);
}
return concurrent_cost;
}
double CMSAdaptiveSizePolicy::concurrent_collection_time() {
double latest_cms_sum_concurrent_phases_time_secs =
_latest_cms_concurrent_marking_time_secs +
_latest_cms_concurrent_precleaning_time_secs +
_latest_cms_concurrent_sweeping_time_secs;
return latest_cms_sum_concurrent_phases_time_secs;
}
double CMSAdaptiveSizePolicy::scaled_concurrent_collection_time() {
// When the precleaning and sweeping phases use multiple
// threads, change one_processor_fraction to
// concurrent_processor_fraction().
double one_processor_fraction = 1.0 / ((double) processor_count());
double latest_cms_sum_concurrent_phases_time_secs =
_latest_cms_concurrent_marking_time_secs * concurrent_processor_fraction() +
_latest_cms_concurrent_precleaning_time_secs * one_processor_fraction +
_latest_cms_concurrent_sweeping_time_secs * one_processor_fraction ;
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print_cr(
"\nCMSAdaptiveSizePolicy::scaled_concurrent_collection_time "
"_latest_cms_concurrent_marking_time_secs %f "
"_latest_cms_concurrent_precleaning_time_secs %f "
"_latest_cms_concurrent_sweeping_time_secs %f "
"concurrent_processor_fraction %f "
"latest_cms_sum_concurrent_phases_time_secs %f ",
_latest_cms_concurrent_marking_time_secs,
_latest_cms_concurrent_precleaning_time_secs,
_latest_cms_concurrent_sweeping_time_secs,
concurrent_processor_fraction(),
latest_cms_sum_concurrent_phases_time_secs);
}
return latest_cms_sum_concurrent_phases_time_secs;
}
void CMSAdaptiveSizePolicy::update_minor_pause_old_estimator(
double minor_pause_in_ms) {
// Get the equivalent of the free space
// that is available for promotions in the CMS generation
// and use that to update _minor_pause_old_estimator
// Don't implement this until it is needed. A warning is
// printed if _minor_pause_old_estimator is used.
}
void CMSAdaptiveSizePolicy::concurrent_marking_begin() {
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print(" ");
gclog_or_tty->stamp();
gclog_or_tty->print(": concurrent_marking_begin ");
}
// Update the interval time
_concurrent_timer.stop();
_latest_cms_collection_end_to_collection_start_secs = _concurrent_timer.seconds();
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print_cr("CMSAdaptiveSizePolicy::concurrent_marking_begin: "
"mutator time %f", _latest_cms_collection_end_to_collection_start_secs);
}
_concurrent_timer.reset();
_concurrent_timer.start();
}
void CMSAdaptiveSizePolicy::concurrent_marking_end() {
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->stamp();
gclog_or_tty->print_cr("CMSAdaptiveSizePolicy::concurrent_marking_end()");
}
_concurrent_timer.stop();
_latest_cms_concurrent_marking_time_secs = _concurrent_timer.seconds();
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print_cr("\n CMSAdaptiveSizePolicy::concurrent_marking_end"
":concurrent marking time (s) %f",
_latest_cms_concurrent_marking_time_secs);
}
}
void CMSAdaptiveSizePolicy::concurrent_precleaning_begin() {
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->stamp();
gclog_or_tty->print_cr(
"CMSAdaptiveSizePolicy::concurrent_precleaning_begin()");
}
_concurrent_timer.reset();
_concurrent_timer.start();
}
void CMSAdaptiveSizePolicy::concurrent_precleaning_end() {
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->stamp();
gclog_or_tty->print_cr("CMSAdaptiveSizePolicy::concurrent_precleaning_end()");
}
_concurrent_timer.stop();
// May be set again by a second call during the same collection.
_latest_cms_concurrent_precleaning_time_secs = _concurrent_timer.seconds();
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print_cr("\n CMSAdaptiveSizePolicy::concurrent_precleaning_end"
":concurrent precleaning time (s) %f",
_latest_cms_concurrent_precleaning_time_secs);
}
}
void CMSAdaptiveSizePolicy::concurrent_sweeping_begin() {
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->stamp();
gclog_or_tty->print_cr(
"CMSAdaptiveSizePolicy::concurrent_sweeping_begin()");
}
_concurrent_timer.reset();
_concurrent_timer.start();
}
void CMSAdaptiveSizePolicy::concurrent_sweeping_end() {
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->stamp();
gclog_or_tty->print_cr("CMSAdaptiveSizePolicy::concurrent_sweeping_end()");
}
_concurrent_timer.stop();
_latest_cms_concurrent_sweeping_time_secs = _concurrent_timer.seconds();
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print_cr("\n CMSAdaptiveSizePolicy::concurrent_sweeping_end"
":concurrent sweeping time (s) %f",
_latest_cms_concurrent_sweeping_time_secs);
}
}
void CMSAdaptiveSizePolicy::concurrent_phases_end(GCCause::Cause gc_cause,
size_t cur_eden,
size_t cur_promo) {
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print(" ");
gclog_or_tty->stamp();
gclog_or_tty->print(": concurrent_phases_end ");
}
// Update the concurrent timer
_concurrent_timer.stop();
if (gc_cause != GCCause::_java_lang_system_gc ||
UseAdaptiveSizePolicyWithSystemGC) {
avg_cms_free()->sample(cur_promo);
double latest_cms_sum_concurrent_phases_time_secs =
concurrent_collection_time();
_avg_concurrent_time->sample(latest_cms_sum_concurrent_phases_time_secs);
// Cost of collection (unit-less)
// Total interval for collection. May not be valid. Tests
// below determine whether to use this.
//
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print_cr("\nCMSAdaptiveSizePolicy::concurrent_phases_end \n"
"_latest_cms_reset_end_to_initial_mark_start_secs %f \n"
"_latest_cms_initial_mark_start_to_end_time_secs %f \n"
"_latest_cms_remark_start_to_end_time_secs %f \n"
"_latest_cms_concurrent_marking_time_secs %f \n"
"_latest_cms_concurrent_precleaning_time_secs %f \n"
"_latest_cms_concurrent_sweeping_time_secs %f \n"
"latest_cms_sum_concurrent_phases_time_secs %f \n"
"_latest_cms_collection_end_to_collection_start_secs %f \n"
"concurrent_processor_fraction %f",
_latest_cms_reset_end_to_initial_mark_start_secs,
_latest_cms_initial_mark_start_to_end_time_secs,
_latest_cms_remark_start_to_end_time_secs,
_latest_cms_concurrent_marking_time_secs,
_latest_cms_concurrent_precleaning_time_secs,
_latest_cms_concurrent_sweeping_time_secs,
latest_cms_sum_concurrent_phases_time_secs,
_latest_cms_collection_end_to_collection_start_secs,
concurrent_processor_fraction());
}
double interval_in_seconds =
_latest_cms_initial_mark_start_to_end_time_secs +
_latest_cms_remark_start_to_end_time_secs +
latest_cms_sum_concurrent_phases_time_secs +
_latest_cms_collection_end_to_collection_start_secs;
assert(interval_in_seconds >= 0.0,
"Bad interval between cms collections");
// Sample for performance counter
avg_concurrent_interval()->sample(interval_in_seconds);
// STW costs (initial and remark pauses)
// Cost of collection (unit-less)
assert(_latest_cms_initial_mark_start_to_end_time_secs >= 0.0,
"Bad initial mark pause");
assert(_latest_cms_remark_start_to_end_time_secs >= 0.0,
"Bad remark pause");
double STW_time_in_seconds =
_latest_cms_initial_mark_start_to_end_time_secs +
_latest_cms_remark_start_to_end_time_secs;
double STW_collection_cost = 0.0;
if (interval_in_seconds > 0.0) {
// cost for the STW phases of the concurrent collection.
STW_collection_cost = STW_time_in_seconds / interval_in_seconds;
avg_cms_STW_gc_cost()->sample(STW_collection_cost);
}
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print("cmsAdaptiveSizePolicy::STW_collection_end: "
"STW gc cost: %f average: %f", STW_collection_cost,
avg_cms_STW_gc_cost()->average());
gclog_or_tty->print_cr(" STW pause: %f (ms) STW period %f (ms)",
(double) STW_time_in_seconds * MILLIUNITS,
(double) interval_in_seconds * MILLIUNITS);
}
double concurrent_cost = 0.0;
if (latest_cms_sum_concurrent_phases_time_secs > 0.0) {
concurrent_cost = concurrent_collection_cost(interval_in_seconds);
avg_concurrent_gc_cost()->sample(concurrent_cost);
// Average this ms cost into all the other types gc costs
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print("cmsAdaptiveSizePolicy::concurrent_phases_end: "
"concurrent gc cost: %f average: %f",
concurrent_cost,
_avg_concurrent_gc_cost->average());
gclog_or_tty->print_cr(" concurrent time: %f (ms) cms period %f (ms)"
" processor fraction: %f",
latest_cms_sum_concurrent_phases_time_secs * MILLIUNITS,
interval_in_seconds * MILLIUNITS,
concurrent_processor_fraction());
}
}
double total_collection_cost = STW_collection_cost + concurrent_cost;
avg_major_gc_cost()->sample(total_collection_cost);
// Gather information for estimating future behavior
double initial_pause_in_ms = _latest_cms_initial_mark_start_to_end_time_secs * MILLIUNITS;
double remark_pause_in_ms = _latest_cms_remark_start_to_end_time_secs * MILLIUNITS;
double cur_promo_size_in_mbytes = ((double)cur_promo)/((double)M);
initial_pause_old_estimator()->update(cur_promo_size_in_mbytes,
initial_pause_in_ms);
remark_pause_old_estimator()->update(cur_promo_size_in_mbytes,
remark_pause_in_ms);
major_collection_estimator()->update(cur_promo_size_in_mbytes,
total_collection_cost);
// This estimate uses the average eden size. It could also
// have used the latest eden size. Which is better?
double cur_eden_size_in_mbytes = ((double)cur_eden)/((double) M);
initial_pause_young_estimator()->update(cur_eden_size_in_mbytes,
initial_pause_in_ms);
remark_pause_young_estimator()->update(cur_eden_size_in_mbytes,
remark_pause_in_ms);
}
clear_internal_time_intervals();
set_first_after_collection();
// The concurrent phases keeps track of it's own mutator interval
// with this timer. This allows the stop-the-world phase to
// be included in the mutator time so that the stop-the-world time
// is not double counted. Reset and start it.
_concurrent_timer.reset();
_concurrent_timer.start();
// The mutator time between STW phases does not include the
// concurrent collection time.
_STW_timer.reset();
_STW_timer.start();
}
void CMSAdaptiveSizePolicy::checkpoint_roots_initial_begin() {
// Update the interval time
_STW_timer.stop();
_latest_cms_reset_end_to_initial_mark_start_secs = _STW_timer.seconds();
// Reset for the initial mark
_STW_timer.reset();
_STW_timer.start();
}
void CMSAdaptiveSizePolicy::checkpoint_roots_initial_end(
GCCause::Cause gc_cause) {
_STW_timer.stop();
if (gc_cause != GCCause::_java_lang_system_gc ||
UseAdaptiveSizePolicyWithSystemGC) {
_latest_cms_initial_mark_start_to_end_time_secs = _STW_timer.seconds();
avg_initial_pause()->sample(_latest_cms_initial_mark_start_to_end_time_secs);
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print(
"cmsAdaptiveSizePolicy::checkpoint_roots_initial_end: "
"initial pause: %f ", _latest_cms_initial_mark_start_to_end_time_secs);
}
}
_STW_timer.reset();
_STW_timer.start();
}
void CMSAdaptiveSizePolicy::checkpoint_roots_final_begin() {
_STW_timer.stop();
_latest_cms_initial_mark_end_to_remark_start_secs = _STW_timer.seconds();
// Start accumumlating time for the remark in the STW timer.
_STW_timer.reset();
_STW_timer.start();
}
void CMSAdaptiveSizePolicy::checkpoint_roots_final_end(
GCCause::Cause gc_cause) {
_STW_timer.stop();
if (gc_cause != GCCause::_java_lang_system_gc ||
UseAdaptiveSizePolicyWithSystemGC) {
// Total initial mark pause + remark pause.
_latest_cms_remark_start_to_end_time_secs = _STW_timer.seconds();
double STW_time_in_seconds = _latest_cms_initial_mark_start_to_end_time_secs +
_latest_cms_remark_start_to_end_time_secs;
double STW_time_in_ms = STW_time_in_seconds * MILLIUNITS;
avg_remark_pause()->sample(_latest_cms_remark_start_to_end_time_secs);
// Sample total for initial mark + remark
avg_cms_STW_time()->sample(STW_time_in_seconds);
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print("cmsAdaptiveSizePolicy::checkpoint_roots_final_end: "
"remark pause: %f", _latest_cms_remark_start_to_end_time_secs);
}
}
// Don't start the STW times here because the concurrent
// sweep and reset has not happened.
// Keep the old comment above in case I don't understand
// what is going on but now
// Start the STW timer because it is used by ms_collection_begin()
// and ms_collection_end() to get the sweep time if a MS is being
// done in the foreground.
_STW_timer.reset();
_STW_timer.start();
}
void CMSAdaptiveSizePolicy::msc_collection_begin() {
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print(" ");
gclog_or_tty->stamp();
gclog_or_tty->print(": msc_collection_begin ");
}
_STW_timer.stop();
_latest_cms_msc_end_to_msc_start_time_secs = _STW_timer.seconds();
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print_cr("CMSAdaptiveSizePolicy::msc_collection_begin: "
"mutator time %f",
_latest_cms_msc_end_to_msc_start_time_secs);
}
avg_msc_interval()->sample(_latest_cms_msc_end_to_msc_start_time_secs);
_STW_timer.reset();
_STW_timer.start();
}
void CMSAdaptiveSizePolicy::msc_collection_end(GCCause::Cause gc_cause) {
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print(" ");
gclog_or_tty->stamp();
gclog_or_tty->print(": msc_collection_end ");
}
_STW_timer.stop();
if (gc_cause != GCCause::_java_lang_system_gc ||
UseAdaptiveSizePolicyWithSystemGC) {
double msc_pause_in_seconds = _STW_timer.seconds();
if ((_latest_cms_msc_end_to_msc_start_time_secs > 0.0) &&
(msc_pause_in_seconds > 0.0)) {
avg_msc_pause()->sample(msc_pause_in_seconds);
double mutator_time_in_seconds = 0.0;
if (_latest_cms_collection_end_to_collection_start_secs == 0.0) {
// This assertion may fail because of time stamp gradularity.
// Comment it out and investiage it at a later time. The large
// time stamp granularity occurs on some older linux systems.
#ifndef CLOCK_GRANULARITY_TOO_LARGE
assert((_latest_cms_concurrent_marking_time_secs == 0.0) &&
(_latest_cms_concurrent_precleaning_time_secs == 0.0) &&
(_latest_cms_concurrent_sweeping_time_secs == 0.0),
"There should not be any concurrent time");
#endif
// A concurrent collection did not start. Mutator time
// between collections comes from the STW MSC timer.
mutator_time_in_seconds = _latest_cms_msc_end_to_msc_start_time_secs;
} else {
// The concurrent collection did start so count the mutator
// time to the start of the concurrent collection. In this
// case the _latest_cms_msc_end_to_msc_start_time_secs measures
// the time between the initial mark or remark and the
// start of the MSC. That has no real meaning.
mutator_time_in_seconds = _latest_cms_collection_end_to_collection_start_secs;
}
double latest_cms_sum_concurrent_phases_time_secs =
concurrent_collection_time();
double interval_in_seconds =
mutator_time_in_seconds +
_latest_cms_initial_mark_start_to_end_time_secs +
_latest_cms_remark_start_to_end_time_secs +
latest_cms_sum_concurrent_phases_time_secs +
msc_pause_in_seconds;
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print_cr(" interval_in_seconds %f \n"
" mutator_time_in_seconds %f \n"
" _latest_cms_initial_mark_start_to_end_time_secs %f\n"
" _latest_cms_remark_start_to_end_time_secs %f\n"
" latest_cms_sum_concurrent_phases_time_secs %f\n"
" msc_pause_in_seconds %f\n",
interval_in_seconds,
mutator_time_in_seconds,
_latest_cms_initial_mark_start_to_end_time_secs,
_latest_cms_remark_start_to_end_time_secs,
latest_cms_sum_concurrent_phases_time_secs,
msc_pause_in_seconds);
}
// The concurrent cost is wasted cost but it should be
// included.
double concurrent_cost = concurrent_collection_cost(interval_in_seconds);
// Initial mark and remark, also wasted.
double STW_time_in_seconds = _latest_cms_initial_mark_start_to_end_time_secs +
_latest_cms_remark_start_to_end_time_secs;
double STW_collection_cost =
collection_cost(STW_time_in_seconds, interval_in_seconds) +
concurrent_cost;
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print_cr(" msc_collection_end:\n"
"_latest_cms_collection_end_to_collection_start_secs %f\n"
"_latest_cms_msc_end_to_msc_start_time_secs %f\n"
"_latest_cms_initial_mark_start_to_end_time_secs %f\n"
"_latest_cms_remark_start_to_end_time_secs %f\n"
"latest_cms_sum_concurrent_phases_time_secs %f\n",
_latest_cms_collection_end_to_collection_start_secs,
_latest_cms_msc_end_to_msc_start_time_secs,
_latest_cms_initial_mark_start_to_end_time_secs,
_latest_cms_remark_start_to_end_time_secs,
latest_cms_sum_concurrent_phases_time_secs);
gclog_or_tty->print_cr(" msc_collection_end: \n"
"latest_cms_sum_concurrent_phases_time_secs %f\n"
"STW_time_in_seconds %f\n"
"msc_pause_in_seconds %f\n",
latest_cms_sum_concurrent_phases_time_secs,
STW_time_in_seconds,
msc_pause_in_seconds);
}
double cost = concurrent_cost + STW_collection_cost +
collection_cost(msc_pause_in_seconds, interval_in_seconds);
_avg_msc_gc_cost->sample(cost);
// Average this ms cost into all the other types gc costs
avg_major_gc_cost()->sample(cost);
// Sample for performance counter
_avg_msc_interval->sample(interval_in_seconds);
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print("cmsAdaptiveSizePolicy::msc_collection_end: "
"MSC gc cost: %f average: %f", cost,
_avg_msc_gc_cost->average());
double msc_pause_in_ms = msc_pause_in_seconds * MILLIUNITS;
gclog_or_tty->print_cr(" MSC pause: %f (ms) MSC period %f (ms)",
msc_pause_in_ms, (double) interval_in_seconds * MILLIUNITS);
}
}
}
clear_internal_time_intervals();
// Can this call be put into the epilogue?
set_first_after_collection();
// The concurrent phases keeps track of it's own mutator interval
// with this timer. This allows the stop-the-world phase to
// be included in the mutator time so that the stop-the-world time
// is not double counted. Reset and start it.
_concurrent_timer.stop();
_concurrent_timer.reset();
_concurrent_timer.start();
_STW_timer.reset();
_STW_timer.start();
}
void CMSAdaptiveSizePolicy::ms_collection_begin() {
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print(" ");
gclog_or_tty->stamp();
gclog_or_tty->print(": ms_collection_begin ");
}
_STW_timer.stop();
_latest_cms_ms_end_to_ms_start = _STW_timer.seconds();
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print_cr("CMSAdaptiveSizePolicy::ms_collection_begin: "
"mutator time %f",
_latest_cms_ms_end_to_ms_start);
}
avg_ms_interval()->sample(_STW_timer.seconds());
_STW_timer.reset();
_STW_timer.start();
}
void CMSAdaptiveSizePolicy::ms_collection_end(GCCause::Cause gc_cause) {
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print(" ");
gclog_or_tty->stamp();
gclog_or_tty->print(": ms_collection_end ");
}
_STW_timer.stop();
if (gc_cause != GCCause::_java_lang_system_gc ||
UseAdaptiveSizePolicyWithSystemGC) {
// The MS collection is a foreground collection that does all
// the parts of a mostly concurrent collection.
//
// For this collection include the cost of the
// initial mark
// remark
// all concurrent time (scaled down by the
// concurrent_processor_fraction). Some
// may be zero if the baton was passed before
// it was reached.
// concurrent marking
// sweeping
// resetting
// STW after baton was passed (STW_in_foreground_in_seconds)
double STW_in_foreground_in_seconds = _STW_timer.seconds();
double latest_cms_sum_concurrent_phases_time_secs =
concurrent_collection_time();
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print_cr("\nCMSAdaptiveSizePolicy::ms_collecton_end "
"STW_in_foreground_in_seconds %f "
"_latest_cms_initial_mark_start_to_end_time_secs %f "
"_latest_cms_remark_start_to_end_time_secs %f "
"latest_cms_sum_concurrent_phases_time_secs %f "
"_latest_cms_ms_marking_start_to_end_time_secs %f "
"_latest_cms_ms_end_to_ms_start %f",
STW_in_foreground_in_seconds,
_latest_cms_initial_mark_start_to_end_time_secs,
_latest_cms_remark_start_to_end_time_secs,
latest_cms_sum_concurrent_phases_time_secs,
_latest_cms_ms_marking_start_to_end_time_secs,
_latest_cms_ms_end_to_ms_start);
}
double STW_marking_in_seconds = _latest_cms_initial_mark_start_to_end_time_secs +
_latest_cms_remark_start_to_end_time_secs;
#ifndef CLOCK_GRANULARITY_TOO_LARGE
assert(_latest_cms_ms_marking_start_to_end_time_secs == 0.0 ||
latest_cms_sum_concurrent_phases_time_secs == 0.0,
"marking done twice?");
#endif
double ms_time_in_seconds = STW_marking_in_seconds +
STW_in_foreground_in_seconds +
_latest_cms_ms_marking_start_to_end_time_secs +
scaled_concurrent_collection_time();
avg_ms_pause()->sample(ms_time_in_seconds);
// Use the STW costs from the initial mark and remark plus
// the cost of the concurrent phase to calculate a
// collection cost.
double cost = 0.0;
if ((_latest_cms_ms_end_to_ms_start > 0.0) &&
(ms_time_in_seconds > 0.0)) {
double interval_in_seconds =
_latest_cms_ms_end_to_ms_start + ms_time_in_seconds;
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print_cr("\n ms_time_in_seconds %f "
"latest_cms_sum_concurrent_phases_time_secs %f "
"interval_in_seconds %f",
ms_time_in_seconds,
latest_cms_sum_concurrent_phases_time_secs,
interval_in_seconds);
}
cost = collection_cost(ms_time_in_seconds, interval_in_seconds);
_avg_ms_gc_cost->sample(cost);
// Average this ms cost into all the other types gc costs
avg_major_gc_cost()->sample(cost);
// Sample for performance counter
_avg_ms_interval->sample(interval_in_seconds);
}
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print("cmsAdaptiveSizePolicy::ms_collection_end: "
"MS gc cost: %f average: %f", cost, _avg_ms_gc_cost->average());
double ms_time_in_ms = ms_time_in_seconds * MILLIUNITS;
gclog_or_tty->print_cr(" MS pause: %f (ms) MS period %f (ms)",
ms_time_in_ms,
_latest_cms_ms_end_to_ms_start * MILLIUNITS);
}
}
// Consider putting this code (here to end) into a
// method for convenience.
clear_internal_time_intervals();
set_first_after_collection();
// The concurrent phases keeps track of it's own mutator interval
// with this timer. This allows the stop-the-world phase to
// be included in the mutator time so that the stop-the-world time
// is not double counted. Reset and start it.
_concurrent_timer.stop();
_concurrent_timer.reset();
_concurrent_timer.start();
_STW_timer.reset();
_STW_timer.start();
}
void CMSAdaptiveSizePolicy::clear_internal_time_intervals() {
_latest_cms_reset_end_to_initial_mark_start_secs = 0.0;
_latest_cms_initial_mark_end_to_remark_start_secs = 0.0;
_latest_cms_collection_end_to_collection_start_secs = 0.0;
_latest_cms_concurrent_marking_time_secs = 0.0;
_latest_cms_concurrent_precleaning_time_secs = 0.0;
_latest_cms_concurrent_sweeping_time_secs = 0.0;
_latest_cms_msc_end_to_msc_start_time_secs = 0.0;
_latest_cms_ms_end_to_ms_start = 0.0;
_latest_cms_remark_start_to_end_time_secs = 0.0;
_latest_cms_initial_mark_start_to_end_time_secs = 0.0;
_latest_cms_ms_marking_start_to_end_time_secs = 0.0;
}
void CMSAdaptiveSizePolicy::clear_generation_free_space_flags() {
AdaptiveSizePolicy::clear_generation_free_space_flags();
set_change_young_gen_for_maj_pauses(0);
}
void CMSAdaptiveSizePolicy::concurrent_phases_resume() {
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->stamp();
gclog_or_tty->print_cr("CMSAdaptiveSizePolicy::concurrent_phases_resume()");
}
_concurrent_timer.start();
}
double CMSAdaptiveSizePolicy::time_since_major_gc() const {
_concurrent_timer.stop();
double time_since_cms_gc = _concurrent_timer.seconds();
_concurrent_timer.start();
_STW_timer.stop();
double time_since_STW_gc = _STW_timer.seconds();
_STW_timer.start();
return MIN2(time_since_cms_gc, time_since_STW_gc);
}
double CMSAdaptiveSizePolicy::major_gc_interval_average_for_decay() const {
double cms_interval = _avg_concurrent_interval->average();
double msc_interval = _avg_msc_interval->average();
double ms_interval = _avg_ms_interval->average();
return MAX3(cms_interval, msc_interval, ms_interval);
}
double CMSAdaptiveSizePolicy::cms_gc_cost() const {
return avg_major_gc_cost()->average();
}
void CMSAdaptiveSizePolicy::ms_collection_marking_begin() {
_STW_timer.stop();
// Start accumumlating time for the marking in the STW timer.
_STW_timer.reset();
_STW_timer.start();
}
void CMSAdaptiveSizePolicy::ms_collection_marking_end(
GCCause::Cause gc_cause) {
_STW_timer.stop();
if (gc_cause != GCCause::_java_lang_system_gc ||
UseAdaptiveSizePolicyWithSystemGC) {
_latest_cms_ms_marking_start_to_end_time_secs = _STW_timer.seconds();
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print_cr("CMSAdaptiveSizePolicy::"
"msc_collection_marking_end: mutator time %f",
_latest_cms_ms_marking_start_to_end_time_secs);
}
}
_STW_timer.reset();
_STW_timer.start();
}
double CMSAdaptiveSizePolicy::gc_cost() const {
double cms_gen_cost = cms_gc_cost();
double result = MIN2(1.0, minor_gc_cost() + cms_gen_cost);
assert(result >= 0.0, "Both minor and major costs are non-negative");
return result;
}
// Cost of collection (unit-less)
double CMSAdaptiveSizePolicy::collection_cost(double pause_in_seconds,
double interval_in_seconds) {
// Cost of collection (unit-less)
double cost = 0.0;
if ((interval_in_seconds > 0.0) &&
(pause_in_seconds > 0.0)) {
cost =
pause_in_seconds / interval_in_seconds;
}
return cost;
}
size_t CMSAdaptiveSizePolicy::adjust_eden_for_pause_time(size_t cur_eden) {
size_t change = 0;
size_t desired_eden = cur_eden;
// reduce eden size
change = eden_decrement_aligned_down(cur_eden);
desired_eden = cur_eden - change;
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print_cr(
"CMSAdaptiveSizePolicy::adjust_eden_for_pause_time "
"adjusting eden for pause time. "
" starting eden size " SIZE_FORMAT
" reduced eden size " SIZE_FORMAT
" eden delta " SIZE_FORMAT,
cur_eden, desired_eden, change);
}
return desired_eden;
}
size_t CMSAdaptiveSizePolicy::adjust_eden_for_throughput(size_t cur_eden) {
size_t desired_eden = cur_eden;
set_change_young_gen_for_throughput(increase_young_gen_for_througput_true);
size_t change = eden_increment_aligned_up(cur_eden);
size_t scaled_change = scale_by_gen_gc_cost(change, minor_gc_cost());
if (cur_eden + scaled_change > cur_eden) {
desired_eden = cur_eden + scaled_change;
}
_young_gen_change_for_minor_throughput++;
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print_cr(
"CMSAdaptiveSizePolicy::adjust_eden_for_throughput "
"adjusting eden for throughput. "
" starting eden size " SIZE_FORMAT
" increased eden size " SIZE_FORMAT
" eden delta " SIZE_FORMAT,
cur_eden, desired_eden, scaled_change);
}
return desired_eden;
}
size_t CMSAdaptiveSizePolicy::adjust_eden_for_footprint(size_t cur_eden) {
set_decrease_for_footprint(decrease_young_gen_for_footprint_true);
size_t change = eden_decrement(cur_eden);
size_t desired_eden_size = cur_eden - change;
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print_cr(
"CMSAdaptiveSizePolicy::adjust_eden_for_footprint "
"adjusting eden for footprint. "
" starting eden size " SIZE_FORMAT
" reduced eden size " SIZE_FORMAT
" eden delta " SIZE_FORMAT,
cur_eden, desired_eden_size, change);
}
return desired_eden_size;
}
// The eden and promo versions should be combined if possible.
// They are the same except that the sizes of the decrement
// and increment are different for eden and promo.
size_t CMSAdaptiveSizePolicy::eden_decrement_aligned_down(size_t cur_eden) {
size_t delta = eden_decrement(cur_eden);
return align_size_down(delta, generation_alignment());
}
size_t CMSAdaptiveSizePolicy::eden_increment_aligned_up(size_t cur_eden) {
size_t delta = eden_increment(cur_eden);
return align_size_up(delta, generation_alignment());
}
size_t CMSAdaptiveSizePolicy::promo_decrement_aligned_down(size_t cur_promo) {
size_t delta = promo_decrement(cur_promo);
return align_size_down(delta, generation_alignment());
}
size_t CMSAdaptiveSizePolicy::promo_increment_aligned_up(size_t cur_promo) {
size_t delta = promo_increment(cur_promo);
return align_size_up(delta, generation_alignment());
}
void CMSAdaptiveSizePolicy::compute_young_generation_free_space(size_t cur_eden,
size_t max_eden_size)
{
size_t desired_eden_size = cur_eden;
size_t eden_limit = max_eden_size;
// Printout input
if (PrintGC && PrintAdaptiveSizePolicy) {
gclog_or_tty->print_cr(
"CMSAdaptiveSizePolicy::compute_young_generation_free_space: "
"cur_eden " SIZE_FORMAT,
cur_eden);
}
// Used for diagnostics
clear_generation_free_space_flags();
if (_avg_minor_pause->padded_average() > gc_pause_goal_sec()) {
if (minor_pause_young_estimator()->decrement_will_decrease()) {
// If the minor pause is too long, shrink the young gen.
set_change_young_gen_for_min_pauses(
decrease_young_gen_for_min_pauses_true);
desired_eden_size = adjust_eden_for_pause_time(desired_eden_size);
}
} else if ((avg_remark_pause()->padded_average() > gc_pause_goal_sec()) ||
(avg_initial_pause()->padded_average() > gc_pause_goal_sec())) {
// The remark or initial pauses are not meeting the goal. Should
// the generation be shrunk?
if (get_and_clear_first_after_collection() &&
((avg_remark_pause()->padded_average() > gc_pause_goal_sec() &&
remark_pause_young_estimator()->decrement_will_decrease()) ||
(avg_initial_pause()->padded_average() > gc_pause_goal_sec() &&
initial_pause_young_estimator()->decrement_will_decrease()))) {
set_change_young_gen_for_maj_pauses(
decrease_young_gen_for_maj_pauses_true);
// If the remark or initial pause is too long and this is the
// first young gen collection after a cms collection, shrink
// the young gen.
desired_eden_size = adjust_eden_for_pause_time(desired_eden_size);
}
// If not the first young gen collection after a cms collection,
// don't do anything. In this case an adjustment has already
// been made and the results of the adjustment has not yet been
// measured.
} else if ((minor_gc_cost() >= 0.0) &&
(adjusted_mutator_cost() < _throughput_goal)) {
desired_eden_size = adjust_eden_for_throughput(desired_eden_size);
} else {
desired_eden_size = adjust_eden_for_footprint(desired_eden_size);
}
if (PrintGC && PrintAdaptiveSizePolicy) {
gclog_or_tty->print_cr(
"CMSAdaptiveSizePolicy::compute_young_generation_free_space limits:"
" desired_eden_size: " SIZE_FORMAT
" old_eden_size: " SIZE_FORMAT,
desired_eden_size, cur_eden);
}
set_eden_size(desired_eden_size);
}
size_t CMSAdaptiveSizePolicy::adjust_promo_for_pause_time(size_t cur_promo) {
size_t change = 0;
size_t desired_promo = cur_promo;
// Move this test up to caller like the adjust_eden_for_pause_time()
// call.
if ((AdaptiveSizePausePolicy == 0) &&
((avg_remark_pause()->padded_average() > gc_pause_goal_sec()) ||
(avg_initial_pause()->padded_average() > gc_pause_goal_sec()))) {
set_change_old_gen_for_maj_pauses(decrease_old_gen_for_maj_pauses_true);
change = promo_decrement_aligned_down(cur_promo);
desired_promo = cur_promo - change;
} else if ((AdaptiveSizePausePolicy > 0) &&
(((avg_remark_pause()->padded_average() > gc_pause_goal_sec()) &&
remark_pause_old_estimator()->decrement_will_decrease()) ||
((avg_initial_pause()->padded_average() > gc_pause_goal_sec()) &&
initial_pause_old_estimator()->decrement_will_decrease()))) {
set_change_old_gen_for_maj_pauses(decrease_old_gen_for_maj_pauses_true);
change = promo_decrement_aligned_down(cur_promo);
desired_promo = cur_promo - change;
}
if ((change != 0) &&PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print_cr(
"CMSAdaptiveSizePolicy::adjust_promo_for_pause_time "
"adjusting promo for pause time. "
" starting promo size " SIZE_FORMAT
" reduced promo size " SIZE_FORMAT
" promo delta " SIZE_FORMAT,
cur_promo, desired_promo, change);
}
return desired_promo;
}
// Try to share this with PS.
size_t CMSAdaptiveSizePolicy::scale_by_gen_gc_cost(size_t base_change,
double gen_gc_cost) {
// Calculate the change to use for the tenured gen.
size_t scaled_change = 0;
// Can the increment to the generation be scaled?
if (gc_cost() >= 0.0 && gen_gc_cost >= 0.0) {
double scale_by_ratio = gen_gc_cost / gc_cost();
scaled_change =
(size_t) (scale_by_ratio * (double) base_change);
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print_cr(
"Scaled tenured increment: " SIZE_FORMAT " by %f down to "
SIZE_FORMAT,
base_change, scale_by_ratio, scaled_change);
}
} else if (gen_gc_cost >= 0.0) {
// Scaling is not going to work. If the major gc time is the
// larger than the other GC costs, give it a full increment.
if (gen_gc_cost >= (gc_cost() - gen_gc_cost)) {
scaled_change = base_change;
}
} else {
// Don't expect to get here but it's ok if it does
// in the product build since the delta will be 0
// and nothing will change.
assert(false, "Unexpected value for gc costs");
}
return scaled_change;
}
size_t CMSAdaptiveSizePolicy::adjust_promo_for_throughput(size_t cur_promo) {
size_t desired_promo = cur_promo;
set_change_old_gen_for_throughput(increase_old_gen_for_throughput_true);
size_t change = promo_increment_aligned_up(cur_promo);
size_t scaled_change = scale_by_gen_gc_cost(change, major_gc_cost());
if (cur_promo + scaled_change > cur_promo) {
desired_promo = cur_promo + scaled_change;
}
_old_gen_change_for_major_throughput++;
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print_cr(
"CMSAdaptiveSizePolicy::adjust_promo_for_throughput "
"adjusting promo for throughput. "
" starting promo size " SIZE_FORMAT
" increased promo size " SIZE_FORMAT
" promo delta " SIZE_FORMAT,
cur_promo, desired_promo, scaled_change);
}
return desired_promo;
}
size_t CMSAdaptiveSizePolicy::adjust_promo_for_footprint(size_t cur_promo,
size_t cur_eden) {
set_decrease_for_footprint(decrease_young_gen_for_footprint_true);
size_t change = promo_decrement(cur_promo);
size_t desired_promo_size = cur_promo - change;
if (PrintAdaptiveSizePolicy && Verbose) {
gclog_or_tty->print_cr(
"CMSAdaptiveSizePolicy::adjust_promo_for_footprint "
"adjusting promo for footprint. "
" starting promo size " SIZE_FORMAT
" reduced promo size " SIZE_FORMAT
" promo delta " SIZE_FORMAT,
cur_promo, desired_promo_size, change);
}
return desired_promo_size;
}
void CMSAdaptiveSizePolicy::compute_tenured_generation_free_space(
size_t cur_tenured_free,
size_t max_tenured_available,
size_t cur_eden) {
// This can be bad if the desired value grows/shrinks without
// any connection to the read free space
size_t desired_promo_size = promo_size();
size_t tenured_limit = max_tenured_available;
// Printout input
if (PrintGC && PrintAdaptiveSizePolicy) {
gclog_or_tty->print_cr(
"CMSAdaptiveSizePolicy::compute_tenured_generation_free_space: "
"cur_tenured_free " SIZE_FORMAT
" max_tenured_available " SIZE_FORMAT,
cur_tenured_free, max_tenured_available);
}
// Used for diagnostics
clear_generation_free_space_flags();
set_decide_at_full_gc(decide_at_full_gc_true);
if (avg_remark_pause()->padded_average() > gc_pause_goal_sec() ||
avg_initial_pause()->padded_average() > gc_pause_goal_sec()) {
desired_promo_size = adjust_promo_for_pause_time(cur_tenured_free);
} else if (avg_minor_pause()->padded_average() > gc_pause_goal_sec()) {
// Nothing to do since the minor collections are too large and
// this method only deals with the cms generation.
} else if ((cms_gc_cost() >= 0.0) &&
(adjusted_mutator_cost() < _throughput_goal)) {
desired_promo_size = adjust_promo_for_throughput(cur_tenured_free);
} else {
desired_promo_size = adjust_promo_for_footprint(cur_tenured_free,
cur_eden);
}
if (PrintGC && PrintAdaptiveSizePolicy) {
gclog_or_tty->print_cr(
"CMSAdaptiveSizePolicy::compute_tenured_generation_free_space limits:"
" desired_promo_size: " SIZE_FORMAT
" old_promo_size: " SIZE_FORMAT,
desired_promo_size, cur_tenured_free);
}
set_promo_size(desired_promo_size);
}
int CMSAdaptiveSizePolicy::compute_survivor_space_size_and_threshold(
bool is_survivor_overflow,
int tenuring_threshold,
size_t survivor_limit) {
assert(survivor_limit >= generation_alignment(),
"survivor_limit too small");
assert((size_t)align_size_down(survivor_limit, generation_alignment())
== survivor_limit, "survivor_limit not aligned");
// Change UsePSAdaptiveSurvivorSizePolicy -> UseAdaptiveSurvivorSizePolicy?
if (!UsePSAdaptiveSurvivorSizePolicy ||
!young_gen_policy_is_ready()) {
return tenuring_threshold;
}
// We'll decide whether to increase or decrease the tenuring
// threshold based partly on the newly computed survivor size
// (if we hit the maximum limit allowed, we'll always choose to
// decrement the threshold).
bool incr_tenuring_threshold = false;
bool decr_tenuring_threshold = false;
set_decrement_tenuring_threshold_for_gc_cost(false);
set_increment_tenuring_threshold_for_gc_cost(false);
set_decrement_tenuring_threshold_for_survivor_limit(false);
if (!is_survivor_overflow) {
// Keep running averages on how much survived
// We use the tenuring threshold to equalize the cost of major
// and minor collections.
// ThresholdTolerance is used to indicate how sensitive the
// tenuring threshold is to differences in cost betweent the
// collection types.
// Get the times of interest. This involves a little work, so
// we cache the values here.
const double major_cost = major_gc_cost();
const double minor_cost = minor_gc_cost();
if (minor_cost > major_cost * _threshold_tolerance_percent) {
// Minor times are getting too long; lower the threshold so
// less survives and more is promoted.
decr_tenuring_threshold = true;
set_decrement_tenuring_threshold_for_gc_cost(true);
} else if (major_cost > minor_cost * _threshold_tolerance_percent) {
// Major times are too long, so we want less promotion.
incr_tenuring_threshold = true;
set_increment_tenuring_threshold_for_gc_cost(true);
}
} else {
// Survivor space overflow occurred, so promoted and survived are
// not accurate. We'll make our best guess by combining survived
// and promoted and count them as survivors.
//
// We'll lower the tenuring threshold to see if we can correct
// things. Also, set the survivor size conservatively. We're
// trying to avoid many overflows from occurring if defnew size
// is just too small.
decr_tenuring_threshold = true;
}
// The padded average also maintains a deviation from the average;
// we use this to see how good of an estimate we have of what survived.
// We're trying to pad the survivor size as little as possible without
// overflowing the survivor spaces.
size_t target_size = align_size_up((size_t)_avg_survived->padded_average(),
generation_alignment());
target_size = MAX2(target_size, generation_alignment());
if (target_size > survivor_limit) {
// Target size is bigger than we can handle. Let's also reduce
// the tenuring threshold.
target_size = survivor_limit;
decr_tenuring_threshold = true;
set_decrement_tenuring_threshold_for_survivor_limit(true);
}
// Finally, increment or decrement the tenuring threshold, as decided above.
// We test for decrementing first, as we might have hit the target size
// limit.
if (decr_tenuring_threshold && !(AlwaysTenure || NeverTenure)) {
if (tenuring_threshold > 1) {
tenuring_threshold--;
}
} else if (incr_tenuring_threshold && !(AlwaysTenure || NeverTenure)) {
if (tenuring_threshold < MaxTenuringThreshold) {
tenuring_threshold++;
}
}
// We keep a running average of the amount promoted which is used
// to decide when we should collect the old generation (when
// the amount of old gen free space is less than what we expect to
// promote).
if (PrintAdaptiveSizePolicy) {
// A little more detail if Verbose is on
GenCollectedHeap* gch = GenCollectedHeap::heap();
if (Verbose) {
gclog_or_tty->print( " avg_survived: %f"
" avg_deviation: %f",
_avg_survived->average(),
_avg_survived->deviation());
}
gclog_or_tty->print( " avg_survived_padded_avg: %f",
_avg_survived->padded_average());
if (Verbose) {
gclog_or_tty->print( " avg_promoted_avg: %f"
" avg_promoted_dev: %f",
gch->gc_stats(1)->avg_promoted()->average(),
gch->gc_stats(1)->avg_promoted()->deviation());
}
gclog_or_tty->print( " avg_promoted_padded_avg: %f"
" avg_pretenured_padded_avg: %f"
" tenuring_thresh: %d"
" target_size: " SIZE_FORMAT
" survivor_limit: " SIZE_FORMAT,
gch->gc_stats(1)->avg_promoted()->padded_average(),
_avg_pretenured->padded_average(),
tenuring_threshold, target_size, survivor_limit);
gclog_or_tty->cr();
}
set_survivor_size(target_size);
return tenuring_threshold;
}
bool CMSAdaptiveSizePolicy::get_and_clear_first_after_collection() {
bool result = _first_after_collection;
_first_after_collection = false;
return result;
}
bool CMSAdaptiveSizePolicy::print_adaptive_size_policy_on(
outputStream* st) const {
if (!UseAdaptiveSizePolicy) return false;
GenCollectedHeap* gch = GenCollectedHeap::heap();
Generation* gen0 = gch->get_gen(0);
DefNewGeneration* def_new = gen0->as_DefNewGeneration();
return
AdaptiveSizePolicy::print_adaptive_size_policy_on(
st,
def_new->tenuring_threshold());
}