CS F301-L19-21

8/10/2019 CS F301-L19-21

1/61

BITSPilaniHyderabad Campus

BITS PilaniDr.Aruna Malapati

Asst ProfessorDepartment of CSIS

8/10/2019 CS F301-L19-21

2/61

BITSPilaniHyderabad Campus

Dynamic Memory Management

8/10/2019 CS F301-L19-21

3/61

CS/IS F301 First Semester 2013-14 BITS Pilani, Hyderabad Campus

Object Lifetime and Storage

Typical Program and Data Layout in Memory

Heap Organization

Heap Allocation

Heap allocation problem

Heap Allocation Algorithms

Todays Agenda

8/10/2019 CS F301-L19-21

4/61


Objects (program data and code) have to bestored in memory during their lifetime.

In general there are three types of objects in aprogram

The objects that are alive throughout the execution ofa program

E.g. global variables

The objects that are alive within a routine

E.g. local variables

The objects whose lifetime can be dynamicallychanged.

The objects that are managed by the new/delete constructs.


8/10/2019 CS F301-L19-21

5/61


The three types of objects correspond to three principal storageallocation mechanisms.

Static objectshave an absolute storage address that is retainedthroughout the execution of the program

Global variables and data

Subroutine code and class method code Stack objects are allocated in last-in first-out order, usually in

conjunction with subroutine calls and returns

Actual arguments passed by value to a subroutine

Local variables of a subroutine

Heap objects may be allocated and deallocated at arbitrarytimes, but require an expensive storage management algorithm

Dynamically allocated data in C++.

Java class instances are always stored on the heap


8/10/2019 CS F301-L19-21

6/61


Typical Program and DataLayout in Memory

Program code is at the bottom

of the memory region (code

section)

The code section is protected

from run-time modification by

the OS

Static data objects are stored

in the static region

Stack grows downward

Heap grows upward

heap

stack

static data

code

0000

Upper addr

Virtualmemoryaddress

space

8/10/2019 CS F301-L19-21

7/61CS/IS F301 First Semester 2013-14 BITS Pilani, Hyderabad Campus

Heap is organized as linear array where dynamic elements

are stored.

Three types of memory blocks

Allocated : active blocks used by process

Released : not used by process need to be recycled to be

in free pool

Free : can be allocated to a process based on run timerequest.

Each memory block contains the size information along

with it.

Heap Organization

8/10/2019 CS F301-L19-21


Heap can be modeled as a sequence of quadruples

where an element of the sequence can be modeled as

(allocated,,,)

(released,,,)

(free,,,)

Heap organization

8/10/2019 CS F301-L19-21


Heap Organization

8/10/2019 CS F301-L19-21


Heap is used to store objects who lifetime is dynamic Implicit heap allocation:

Done automatically

Java class instances are placed on the heap

Scripting languages and functional languages make

extensive use of the heap for storing objects

Some procedural languages allow array declarations with

run-time dependent array size

Resizable character strings

Explicit heap allocation:

Statements and/or functions for allocation and deallocation

Malloc/free, new/delete

Heap Allocation

8/10/2019 CS F301-L19-21


Heap is a large block of memory (say N bytes)

Requests for memory of various sizes may arriverandomly: program runs new

Each request may ask for 1 to N bytes

If a request of X bytes is granted, a continuousX bytesin the heap is allocated for the request. The memory willbe used for a while and then return to the system (when

the program runs delete

The problem: how to allocate memory so that as manyrequest can be satisfied as possible.

Heap allocation problem

8/10/2019 CS F301-L19-21


Heap Allocation Algorithms

Heap allocation is performed by searching the heap foravailable free space

For example, suppose we want to allocate a new objectE of 20 bytes, where would it fit?

Deletion of objects leaves free blocks in the heap thatcan be reused

Two questions

How to keep track of free blocks?

How to find the right free block for each request?

Object A Free Object B Object C Free Object D Free30 bytes 8 bytes 10 bytes 24 bytes 24 bytes 8 bytes 20 bytes

8/10/2019 CS F301-L19-21


How to keep track of free blocks? Maintain a linked list of free heap blocks

To satisfy a request (new), search the list to find oneblock whose size is equal to or larger than the

requested size If the size is equal, remove the block from the free list

If the size is larger, modify the size to (sizerequestedsize).

When an object is deleted (freed), the block ofmemory is returned to the heap.

Insert a new block to the free list. If new block can bemerged with its neighboring blocks to be a larger block,merge the blocks.

Heap Allocation Algorithms(contd)

8/10/2019 CS F301-L19-21


How to pick which block to be used for each

request?

There can be many choices. First-fit: select the first block in the list that is large

enough

Best-fit: search the entire list for the smallest free

block that is large enough to hold the object

O(N) for both new and delete operations

Heap Allocation Algorithms(contd)

8/10/2019 CS F301-L19-21


Example: Best Fit

8/10/2019 CS F301-L19-21


Example: First Fit

8/10/2019 CS F301-L19-21


Example: Worst Fit

8/10/2019 CS F301-L19-21


Example: Next Fit

8/10/2019 CS F301-L19-21


8/10/2019 CS F301-L19-21


Manual (for Ex free in C language)

Automatic (Garbage collection)

Deallocation does not require traversal of every memory

allocation.

Fragmentaion

Deallocation of objects

8/10/2019 CS F301-L19-21


Start and stop: Suspends execution of the program

temporarily when it starts execution.

Real time: interleaves the garbage collection with

program execution.

Incremental garbage collection

Continuous garbage collection: collects released memory

immediately

Reference count

Concurrent garbage collection

Hard real time garbage collection

Garbage collectionapproaches

8/10/2019 CS F301-L19-21

23/61


Mark and scan algorithm

Copying garbage collection

Cheneys modified Copying garbage collection Generational garbage collection

Start and stop garbagecollection

8/10/2019 CS F301-L19-21

24/61


Has two phases

Mark

Scan

Mark and scan algorithm

8/10/2019 CS F301-L19-21

25/61


Each cell has a mark bit

Garbage remains unreachable and undetected until

heap is used up; then GC goes to work, while program

execution is suspended

Marking phase

Starting from the roots, set the mark bit on all live

cells

Sweep phase

Return all unmarked cells to the free list

Reset the mark bit on all marked cells

Mark-Sweep GarbageCollection

8/10/2019 CS F301-L19-21

26/61


Mark-Sweep Example

root

set

Heap space

8/10/2019 CS F301-L19-21

27/61


Mark-Sweep Example

root

set

Heap space

8/10/2019 CS F301-L19-21

28/61


Mark-Sweep Example

root

set

Heap space

8/10/2019 CS F301-L19-21

29/61


Mark-Sweep Example

root

set

Heap space

Reset mark bitof marked cells

Free unmarkedcells

8/10/2019 CS F301-L19-21

30/61


Copying garbage collection

Divide heap into two semispaces.Allocate from one space (fromspace) till full.

Copy live data into other space (tospace).

Switch roles of the spaces.

Requires fixing pointers to moved data

(forwarding).

Eliminates fragmentation.

DFS improves locality, while BFS does not

require any extra storage.

8/10/2019 CS F301-L19-21

31/61


Stop and Copy GC: Example

A B C D Froot E

Before collection:

new space

A C F

root

new space

After collection:

free

heappointer

8/10/2019 CS F301-L19-21

32/61


Form an initial queue of objects which can beimmediately reached from the root set.

A "scan" pointer is advanced through the objects location

by location. Every time a pointer into fromspace isencountered, the object the pointer refers to is copied to

the end of the queue.

When the "scan" reaches the end of the queue, all liveobjects have been copied, so the garbage collector is

terminated.

Cheney's Algorithm

8/10/2019 CS F301-L19-21

33/61

8/10/2019 CS F301-L19-21

34/61


Advantages

The allocation of free objects is simple and fast.

This method does not cause memory fragmentation,

even when objects of different sizes are copied.

Optimization

To increase copying collectors efficiency, increase theamount of memory allocated for the heap space to

reduce the number of times the collector is invoked.

Cheney's Algorithm

8/10/2019 CS F301-L19-21

35/61


Generational garbage collection divides the heap into

two or more regions, called generations.

Objects are always allocated in the youngest generation.

The garbage collection algorithm scans the youngest

generation most frequently, and performs scanning of

successive generation more rarely.

Generational GarbageCollection

G G

8/10/2019 CS F301-L19-21

36/61


Object allocation

Generational Garbagecollection

G i l G b

8/10/2019 CS F301-L19-21

37/61


When the eden space fills up, a minor garbage collectionis triggered.


G ti l G b

8/10/2019 CS F301-L19-21

38/61


Referenced objects are moved to the first survivorspace. Unreferenced objects are deleted when the eden

space is cleared.


G ti l G b

8/10/2019 CS F301-L19-21

39/61


At the next minor GC, the same thing happens for theeden space. Unreferenced objects are deleted and

referenced objects are moved to a survivor space.

However, in this case, they are moved to the second

survivor space (S1).

In addition, objects from the last minor GC on the first

survivor space (S0) have their age incremented and get

moved to S1. Once all surviving objects have beenmoved to S1, both S0 and eden are cleared.

Notice we now have differently aged object in the

survivor space.


8/10/2019 CS F301-L19-21

40/61


Object aging

8/10/2019 CS F301-L19-21

41/61


At the next minor GC, the same process repeats.However this time the survivor spaces switch.

Referenced objects are moved to S0. Surviving objects

are aged. Eden and S1 are cleared.

Additional aging

8/10/2019 CS F301-L19-21

42/61


After a minor GC, when aged objects reach a certain agethreshold (8 in this example) they are promoted from

young generation to old generation.

Promotion

8/10/2019 CS F301-L19-21

43/61


As minor GCs continue to occure objects will continue tobe promoted to the old generation space

Promotion

8/10/2019 CS F301-L19-21

44/61


So that pretty much covers the entire process with theyoung generation. Eventually, a major GC will be

performed on the old generation which cleans up and

compacts that space.

GC process summary

I t l G b

8/10/2019 CS F301-L19-21

45/61


Interleaves Partial GC(PGC) with ProgramExecution(PE)

Once GC starts PGC,PE, PGC,PE, PGC,PE, where

PGC is very small time slice.

More memory is recycled than allocated.

After GC is over PE continues execution.

Incremental Garbagecollection

T i l M ki d

8/10/2019 CS F301-L19-21

46/61


Tricolor marking is a method of marking which objectshave been looked at in a collection cycle, and

determining which ones to recycle at the end of the

cycle.

Grey Are ready to be examined by the collector

Are assumed to be in use by the mutator

Black Have already been examined by the collector

Are assumed to be in use by the mutator

White Have not yet been examined by the collector

May or may not be in use by the mutator

Tricolor Marking andCoherence

Continuous GC using

8/10/2019 CS F301-L19-21

47/61


Continuous GC usingreference counting

8/10/2019 CS F301-L19-21

48/61


Periodically copy K (K>1) cells from space to the tospace.

When GC starts the Form space is sealed and memory

allocation is in the tospace.

If m blocks are requested them m*k blocks are copied

fromspace to tospace.

Bakers algorithm

8/10/2019 CS F301-L19-21

49/61


Black:The cells that arecopied formspace to

tospace along with

children.

Gray: The cells that copiedformspace to tospace has

links to children in the

other space.

White: cells are that arereleased and will not be

copied.

Bakers algorithm

8/10/2019 CS F301-L19-21

50/61


The root set is all the data that can be accessed (reached)directly by a program without having to dereference any pointer

Recursively, any object whose reference is stored in a field of a

member of the root set is also reachable

New objects are introduced through object allocations and add

to the set of reachable objects

Parameter passing and assignments can propagate reachability

Assignments and ends of procedures can terminate reachability

Reachability of Objects

8/10/2019 CS F301-L19-21

51/61


Similarly, an object that becomes unreachable can causemore objects to become unreachable

A garbage collector periodically finds all unreachable

objects by one of the two methods. Catch the transitionsas reachable objects become unreachable

Or, periodically locate all reachable objects and infer that

all other objects are unreachable

Reachability of Objects

8/10/2019 CS F301-L19-21

52/61

8/10/2019 CS F301-L19-21

53/61


New object allocation.ref_count=1for the new object

Parameter passing.ref_count++for each object passed

into a procedure

Reference assignments.For u:=v, where u and v are

references, ref_count++for the object *v, and ref_count--for the object *u

Procedure returns. ref_count--for each object pointed to

by the local variables

Transitive loss of reachability.Whenever ref_countof anobject becomes zero, we must also decrement the

ref_countof each object pointed to by a reference within

the object

Maintaining Reference Counts

8/10/2019 CS F301-L19-21

54/61


Reference Count Manipulation

8/10/2019 CS F301-L19-21

55/61



8/10/2019 CS F301-L19-21

56/61


Node D is now unreachable.It is collected too.


Indicated number is

the reference count

Reference Counting GC:

8/10/2019 CS F301-L19-21

57/61


High overhead due to reference maintenance Cannot collect unreachable cyclic data structures (ex:

circularly linked lists), since the reference counts never

become zero

Garbage collection is incrementaloverheads aredistributed to the mutatorsoperations and are spread out

throughout the life time of the mutator

Garbage is collected immediately and hence spaceusage is low

Useful for real-time and interactive applications, where

long and sudden pauses are unacceptable

Reference Counting GC:Advantages and Disadvantages

Unreachable Cyclic Data

8/10/2019 CS F301-L19-21

58/61


Unreachable Cyclic DataStructure

8/10/2019 CS F301-L19-21

59/61


Example

Concurrent garbage

8/10/2019 CS F301-L19-21

60/61


Two conditions to be satisfied The Memory space of the collection process and the

memory allocation process are separated.

The two process are synchronized when working on the

same memory location.

Two process in this process Mutator: allocate or release the memory locations.

Collector: coverts the released memory blocks into free memory

Concurrent garbagecollection

8/10/2019 CS F301-L19-21

61/61

Needs info as to whether a cell is a data or pointer.

Takes 20-30% of execution time.

Start and stop GC has performance issues.

Issues in GC

Documents

CS F301-L19-21