3681N/A * Copyright (c) 2011, 2012, Oracle and/or its affiliates. All rights reserved. 2386N/A * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 2386N/A * This code is free software; you can redistribute it and/or modify it 2386N/A * under the terms of the GNU General Public License version 2 only, as 2386N/A * published by the Free Software Foundation. 2386N/A * This code is distributed in the hope that it will be useful, but WITHOUT 2386N/A * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 2386N/A * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 2386N/A * version 2 for more details (a copy is included in the LICENSE file that 2386N/A * You should have received a copy of the GNU General Public License version 2386N/A * 2 along with this work; if not, write to the Free Software Foundation, 2386N/A * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 2386N/A * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 2816N/A// Class for monitoring logical spaces in G1. It provides data for 2816N/A// both G1's jstat counters as well as G1's memory pools. 2816N/A// G1 splits the heap into heap regions and each heap region belongs 2816N/A// to one of the following categories: 2816N/A// * eden : regions that have been allocated since the last GC 2816N/A// * survivors : regions with objects that survived the last few GCs 2816N/A// * old : long-lived non-humongous regions 2816N/A// * humongous : humongous regions 2816N/A// The combination of eden and survivor regions form the equivalent of 2816N/A// the young generation in the other GCs. The combination of old and 2816N/A// humongous regions form the equivalent of the old generation in the 2816N/A// other GCs. Free regions do not have a good equivalent in the other 2816N/A// GCs given that they can be allocated as any of the other region types. 2816N/A// The monitoring tools expect the heap to contain a number of 2816N/A// generations (young, old, perm) and each generation to contain a 2816N/A// number of spaces (young: eden, survivors, old). Given that G1 does 2816N/A// not maintain those spaces physically (e.g., the set of 2816N/A// non-contiguous eden regions can be considered as a "logical" 2816N/A// space), we'll provide the illusion that those generations and 2816N/A// spaces exist. In reality, each generation and space refers to a set 2816N/A// of heap regions that are potentially non-contiguous. 2816N/A// This class provides interfaces to access the min, current, and max 2816N/A// capacity and current occupancy for each of G1's logical spaces and 2816N/A// generations we expose to the monitoring tools. Also provided are 2816N/A// counters for G1 concurrent collections and stop-the-world full heap 2816N/A// Below is a description of how the various sizes are calculated. 2816N/A// - heap_capacity = current heap capacity (e.g., current committed size) 2816N/A// - young_gen_capacity = current max young gen target capacity 2816N/A// (i.e., young gen target capacity + max allowed expansion capacity) 2816N/A// - survivor_capacity = current survivor region capacity 2816N/A// - eden_capacity = young_gen_capacity - survivor_capacity 2816N/A// - old_capacity = heap_capacity - young_gen_capacity 2816N/A// What we do in the above is to distribute the free regions among 2816N/A// eden_capacity and old_capacity. 2816N/A// - young_gen_used = current young region capacity 2816N/A// - survivor_used = survivor_capacity 2816N/A// - eden_used = young_gen_used - survivor_used 2816N/A// - old_used = overall_used - young_gen_used 2816N/A// Unfortunately, we currently only keep track of the number of 2816N/A// currently allocated young and survivor regions + the overall used 2816N/A// bytes in the heap, so the above can be a little inaccurate. 3115N/A// We set this to 0 for all spaces. 2816N/A// For jstat, we set the max capacity of all spaces to heap_capacity, 3115N/A// given that we don't always have a reasonable upper bound on how big 3115N/A// each space can grow. For the memory pools, we make the max 3115N/A// capacity undefined with the exception of the old memory pool for 3115N/A// which we make the max capacity same as the max heap capacity. 2816N/A// If we had more accurate occupancy / capacity information per 2816N/A// region set the above calculations would be greatly simplified and 2816N/A// We update all the above synchronously and we store the results in 2816N/A// fields so that we just read said fields when needed. A subtle point 2816N/A// is that all the above sizes need to be recalculated when the old 2816N/A// gen changes capacity (after a GC or after a humongous allocation) 2816N/A// but only the eden occupancy changes when a new eden region is 2816N/A// allocated. So, in the latter case we have minimal recalcuation to 2816N/A// do which is important as we want to keep the eden region allocation 2816N/A// path as low-overhead as possible. 2386N/A // jstat performance counters 2986N/A // incremental collections both young and mixed 2386N/A // full stop-the-world collections 2386N/A // young collection set counters. The _eden_counters, 2386N/A // _from_counters, and _to_counters are associated with 2386N/A // this "generational" counter. 2816N/A // old collection set counters. The _old_space_counters 2386N/A // below are associated with this "generational" counter. 2386N/A // Counters for the capacity and used for 2386N/A // the survivor collection (only one, _to_counters, is actively used) 2816N/A // When it's appropriate to recalculate the various sizes (at the 2816N/A // end of a GC, when a new eden region is allocated, etc.) we store 2816N/A // them here so that we can easily report them when needed and not 2816N/A // have to recalculate them every time. 2386N/A // It returns x - y if x > y, 0 otherwise. 2386N/A // As described in the comment above, some of the inputs to the 2386N/A // calculations we have to do are obtained concurrently and hence 2386N/A // may be inconsistent with each other. So, this provides a 2386N/A // defensive way of performing the subtraction and avoids the value 2386N/A // going negative (which would mean a very large result, given that 2386N/A // the parameter are size_t). 2816N/A // Recalculate all the sizes. 2816N/A // Recalculate only what's necessary when a new eden region is allocated. 2816N/A // Unfortunately, the jstat tool assumes that no space has 0 2816N/A // capacity. In our case, given that each space is logical, it's 2816N/A // possible that no regions will be allocated to it, hence to have 0 2816N/A // capacity (e.g., if there are no survivor regions, the survivor 2816N/A // space has 0 capacity). The way we deal with this is to always pad 2816N/A // each capacity value we report to jstat by a very small amount to 2816N/A // make sure that it's never zero. Given that we sometimes have to 2816N/A // report a capacity of a generation that contains several spaces 2816N/A // (e.g., young gen includes one eden, two survivor spaces), the 2816N/A // mult parameter is provided in order to adding the appropriate 2816N/A // padding multiple times so that the capacities add up correctly. 2816N/A // Recalculate all the sizes from scratch and update all the jstat 2816N/A // Recalculate only what's necessary when a new eden region is 2816N/A // allocated and update any jstat counters that need to be updated. 2386N/A // Monitoring support used by 2386N/A#
endif // SHARE_VM_GC_IMPLEMENTATION_G1_G1MONITORINGSUPPORT_HPP