Chapter 12 Memory Management

Data Structures and Algorithms in Java

Chapter 12

Memory Management

Data Structures and Algorithms in Java 2

Objectives

Discuss the following topics: • Memory Management• The Sequential-Fit Methods• The Nonsequential-Fit Methods• Garbage Collection• Case Study: An In-Place Garbage Collector


Memory Management

• The heap is the region of main memory from which portions of memory are dynamically allocated upon request of a program

• The memory manager is responsible for:– The maintenance of free memory blocks– Assigning specific memory blocks to the user

programs – Cleaning memory from unneeded blocks

to return them to the memory pool


Memory Management (continued)

• The memory manager is responsible for:– Scheduling access to shared data, – Moving code and data between main and

secondary memory– Keeping one process away from another

• External fragmentation amounts to the presence of wasted space between allocated segments of memory


Memory Management (continued)

• Internal fragmentation amounts to the presence of unused memory inside the segments


The Sequential-Fit Methods

• In the sequential-fit methods, all available memory blocks are linked, and the list is searched to find a block whose size is larger than or the same as the requested size

• The first-fit algorithm allocates the first block of memory large enough to meet the request

• The best-fit algorithm allocates a block that is closest in size to the request


The Sequential-Fit Methods (continued)

• The worst-fit method finds the largest block on the list so that the remaining part is large enough to be used in later requests

• The next-fit method allocates the next available block that is sufficiently large

• The way the blocks are organized on the list determines how fast the search for an available block succeeds or fails


The Sequential-Fit Methods (continued)

Figure 12-1 Memory allocation using sequential-fit methods


The Nonsequential-Fit Methods

• An adaptive exact-fit technique dynamically creates and adjusts storage block lists that fit the requests exactly

• In adaptive exact-fit, a size-list of block lists of a particular size returned to the memory pool during the last T allocations is maintained

• The exact-fit method disposes of entire block lists if no request comes for a block from this list in the last T allocations


The Nonsequential-Fit Methods (continued)

t = 0;allocate (reqSize)

t++;if a block list b1 with reqSize blocks is on sizeList

lastref(b1) = t;b = head of blocks(b1);if b was the only block accessible from b1

detach b1 from sizeList;else b = search-memory-for-a-block-of(reqSize);dispose of all block lists on sizeList for which t - lastref(b1) < T;return b;


The Nonsequential-Fit Methods (continued)

Figure 12-2 An example configuration of a size-list and heap created by the adaptive exact-fit method


Buddy Systems

• Nonsequential memory management methods or buddy systems assign memory in sequential slices, and divide it into two buddies that are merged whenever possible

• In the buddy system, two buddies are never free• A block can have either a buddy used by the

program or none


Buddy Systems (continued)

• In the binary buddy system each block of memory (except the entire memory) is coupled with a buddy of the same size that participates with the block in reserving and returning chunks of memory



Figure 12-3 Block structure in the binary buddy system



Figure 12-4 Reserving three blocks of memory using the binary buddy system



Figure 12-4 Reserving three blocks of memory using the binary buddy system (continued)



Figure 12-4 Reserving three blocks of memory using the binary buddy system (continued)



Figure 12-5 (a) Returning a block to the pool of blocks, (b) resulting in coalescing one block with its buddy



Figure 12-5 (c) Returning another block leads to two coalescings (continued)



avail[i] = -1 for i = 0, . . . , m-1;avail[m] = first address in memory;

reserveFib(reqSize)availSize = the position of the first Fibonacci number greater than

reqSizefor which avail[availSize] > -1;



Figure 12-6 (a) Splitting a block of size Fib(k) into two buddies using the buddy-bit and the memory-bit



Figure 12-6 (b) Coalescing two buddies utilizing information stored in buddy- and memory-bits



• A weighted buddy system is to decrease the amount of internal fragmentation by allowing more block sizes than in the binary system

• A buddy system that takes a middle course between the binary system and the weighted system is a dual buddy system


Garbage Collection

• A garbage collector is automatically invoked to collect unused memory cells when the program is idle or when memory resources are exhausted

• References to all linked structures currently utilized by the program are stored in a root set, which contains all root pointers


Garbage Collection (continued)

• There are two phases of garbage collection:– The marking phase — to identify all currently

used cells– The reclamation phase — when all unmarked

cells are returned to the memory pool; this phase can also include heap compaction


Mark-and-Sweep

• Memory cells currently in use are marked by:– Traversing each linked structure– Then the memory is swept to glean unused

(garbage) cells that put them together in a memory pool

marking (node)if node is not markedmark node;if node is not an atommarking(head(node));marking(tail(node));


Mark-and-Sweep (continued)

Figure 12-7 An example of execution of the Schorr and Waite algorithm for marking used memory cells



Figure 12-7 An example of execution of the Schorr and Waite algorithm for marking used memory cells (continued)














Space Reclamation

sweep()for each location from the last to the first

if mark(location) is 0 insert location in front of availList;else set mark(location) to 0;


Compaction

Figure 12-8 An example of heap compaction


Copying Methods

• The stop-and-copy algorithm divides the heap into two semispaces, one of which is only used for allocating memory

• Lists can be copied using breadth-first traversal that allows it to combine two tasks: copying lists and updating references

• This algorithm requires no marking phase and no stack


Copying Methods (continued)

Figure 12-9 (a) A situation in the heap before copying the contents of cells in use from semispace1 to semispace2


Copying Methods (continued)

Figure 12-9 (b) the situation right after copying; all used cells are packed contiguously (continued)


Incremental Garbage Collection

• Incremental garbage collectors, whose execution is interleaved with the execution of the program, is desirable for a fast response to a program

• After the collector partially processes some lists, the program can change or mutate those lists using a program called a mutator

• The Baker algorithm uses two semispaces, called fromspace and tospace, which are both active to ensure proper cooperation between the mutator and the collector


Incremental Garbage Collection (continued)

Figure 12-10 A situation in memory (a) before and (b) after allocating a cell with head and tail references referring to cells P and Q in tospace according to the Baker algorithm



Figure 12-10 A situation in memory (a) before and (b) after allocating a cell with head and tail references referring to cells P and Q in tospace according to the Baker algorithm (continued)



• The mutator is preceded by a read barrier, which precludes utilizing references to cells in fromspace

• The generational garbage collection technique divides all allocated cells into at least two generations and focuses its attention on the youngest generation, which generates most of the garbage



Figure 12-11 Changes performed by the Baker algorithm when addresses P and Q refer to cells in fromspace, P to an already copied cell, Q to a cell still in fromspace



Figure 12-11 Changes performed by the Baker algorithm when addresses P and Q refer to cells in fromspace, P to an already copied cell, Q to a cell still in fromspace (continued)



Figure 12-12 A situation in three regions (a) before and (b) after copying reachable cells from region ri to region r’i in the Lieberman-Hewitt technique of generational garbage collection



Figure 12-12 A situation in three regions (a) before and (b) after copying reachable cells from region ri to region r’i in the Lieberman- Hewitt technique of generational garbage collection (continued)


createRootPtr(p,q,r) // Lisp’s consif collector is in the marking phase

mark up to k1 cells;else if collector is in the sweeping phase

sweep up to k2 cells;else if the number of cells on availList is low

push all root pointers onto collector’s stack st;p = first cell on availList;head(p) = q;tail(p) = r;mark p if it is in the unswept portion of heap;

Noncopying Methods


Noncopying Methods (continued)

Figure 12-13 An inconsistency that results if, in Yuasa’s noncopying incremental garbage collector, a stack is not used to record cells possibly unprocessed during the marking phase



Figure 12-13 An inconsistency that results if, in Yuasa’s noncopying incremental garbage collector, a stack is not used to record cells possibly unprocessed during the marking phase (continued)



Figure 12-14 Memory changes during the sweeping phase using Yuasa’s method



Figure 12-14 Memory changes during the sweeping phase using Yuasa’s method (continued)


Case Study: An In-Place Garbage Collector

roots: 1 5 3(0: -1 2 false false 0 0) (1: 5 4 false false 1 4)(2: 0 -1 false false 2 2) (3: 4 -1 true false 130)(4: 1 3 true false 129) (5: -1 1 false false 5 1)freeCells: (0 0 0)(2 2 2)nonFreeCells: (5 5 1)(1 1 4)(4 129)(3 130)


Case Study: An In-Place Garbage Collector (continued)

Figure 12-15 An example of a situation on the heap



Figure 12-15 An example of a situation on the heap (continued)



Figure 12-16 Implementation of an in-place garbage collector



Figure 12-16 Implementation of an in-place garbage collector (continued)












































Summary

• The heap is the region of main memory from which portions of memory are dynamically allocated upon request of a program

• External fragmentation amounts to the presence of wasted space between allocated segments of memory

• Internal fragmentation amounts to the presence of unused memory inside the segments


Summary (continued)

• In the sequential-fit methods, all available memory blocks are linked, and the list is searched to find a block whose size is larger than or the same as the requested size

• An adaptive exact-fit technique dynamically creates and adjusts storage block lists that fit the requests exactly

• Nonsequential memory management methods or buddy systems assign memory in sequential slices, and divide it into two buddies that are merged whenever possible


Summary (continued)

• A garbage collector is automatically invoked to collect unused memory cells when the program is idle or when memory resources are exhausted

• The stop-and-copy algorithm divides the heap into two semispaces, with one only used for allocating memory

• Incremental garbage collectors, whose execution is interleaved with the execution of the program, is desirable for a fast response to a program

Chapter 12 Memory Management

Documents