Memory Management

Jordan University of Science & Technology

CPE 746 Embedded Real-Time Systems

Prepared By: Salam Al-Mandil & Hala Obaidat

Supervised By: Dr. Lo’ai Tawalbeh

Outline Introduction Common Memory Types Composing Memory Memory Hierarchy Caches Application Memory Management Static memory management Dynamic memory management Memory Allocation The problem of fragmentation Memory Protection Recycling techniques

Introduction An embedded system is a special-purpose

computer system designed to perform one or a few dedicated functions, sometimes with real-time computing constraints.

An embedded system is part of a larger system.

Embedded systems often have small memory and are required to run a long time, so memory management is a major concern when developing real-time applications.

Common Memory Types

RAM DRAM: Volatile memory. Address lines are multiplexed. The 1st half is

sent 1st & called the RAS. The 2nd half is sent later & called CAS.

A capacitor and a single transistor / bit => better capacity.

Requires periodical refreshing every 10-100 ms => dynamic.

Cheaper / bit => lower cost. Slower => used for main memory.

Reading DRAM Super cell (2,1)

Step 1(a): Row access strobe (RAS) selects row 2.

cols

rows

RAS = 20 1 2 3

0

1

2

internal row buffer

16 x 8 DRAM chip

3

addr

data

2/

8/

memorycontroller

Step 1(b): Row 2 copied from DRAM array to row buffer.

Reading DRAM Super cell (2,1)

Step 2(a): Column access strobe (CAS) selects column 1.

cols

rows

0 1 2 3

0

1

2

3

internal row buffer

16 x 8 DRAM chip

CAS = 1

addr

data

2/

8/

memorycontroller

Step 2(b): Super cell (2,1) copied from buffer to data lines, and eventually back to the CPU.

super cell (2,1)

super cell (2,1)

To CPU

RAM

SRAM: Volatile memory. Six transistors / bit => lower capacity. No refreshing required => faster &

lower power consumption. More expensive / bit => higher cost. Faster => used in caches.

Some Memory Types ROM: Non-volatile memory. Can be read from but not written to, by

a processor in an embedded system. Traditionally written to, “programmed”,

before inserting to embedded system. Stores constant data needed by system. Horizontal lines = words, vertical lines =

data. Some embedded systems work without

RAM, exclusively on ROM, because their programs and data are rarely changed.

Some Memory Types

Flash Memory: Non-volatile memory. Can be electrically erased & reprogrammed. Used in memory cards, and USB flash drives. It is erased and programmed in large blocks

at once, rather than one word at a time. Examples of applications include PDAs and

laptop computers, digital audio players, digital cameras and mobile phones.

Type Volatile? Writeable? Erase SizeMax Erase

CyclesCost (per

Byte)Speed

SRAM Yes Yes Byte Unlimited Expensive Fast

DRAM Yes Yes Byte Unlimited Moderate Moderate

Masked ROM

No No n/a n/a Inexpensive Fast

PROM No

Once, with a device

programmer

n/a n/a Moderate Fast

EPROM No

Yes, with a device

programmer

Entire Chip

Limited (consult datashe

et)

Moderate Fast

EEPROM No Yes Byte

Limited (consult datashe

et)

Expensive

Fast to read,

slow to erase/wr

ite

Flash No Yes Sector

Limited (consult datashe

et)

Moderate

Fast to read,

slow to erase/wr

ite

NVRAM No Yes Byte Unlimited

Expensive (SRAM

+ battery)

Fast

Composing Memory

When available memory is larger, simply ignore unneeded high-order address bits and higher data lines.

When available memory is smaller, compose several smaller memories into one larger memory:

Connect side-by-side. Connect top to bottom. Combine techniques.

Connect side-by-side

To increase width of words.

…

2m × 3n ROM

2m × n ROM

A0

…

enable 2m × n ROM

…

2m × n ROM

…

Q3n-1 Q2n-1

…

Q0

…

Am

Increase width of words

Connect top to bottom

To increase number of words.

2m+1 × n ROM

2m × n ROM

A0…

enable

…

2m × n ROM

Am-1

Am

1 × 2 decoder

…

…

…

Qn-1 Q0

…

Increase number of words

Combine techniques

To increase number and width of words.

A

enable

outputs

Increase number and width of words

Memory Hierarchy Is an approach for organizing memory

and storage systems. A memory hierarchy is organized into

several levels – each smaller, faster, & more expensive / byte than the next lower level.

For each k, the faster, smaller device at level k serves as a cache for the larger, slower device at level k+1.

Programs tend to access the data at level k more often than they access the data at level k+1.

An Example Memory Hierarchy

registers

on-chip L1cache (SRAM)

main memory(DRAM)

local secondary storage(local disks)

Larger, slower,

and cheaper (per byte)storagedevices

remote secondary storage(distributed file systems, Web servers)

Local disks hold files retrieved from disks on remote network servers.

Main memory holds disk blocks retrieved from local disks.

off-chip L2cache (SRAM)

L1 cache holds cache lines retrieved from the L2 cache memory.

CPU registers hold words retrieved from L1 cache.

L2 cache holds cache lines retrieved from main memory.

L0:

L1:

L2:

L3:

L4:

L5:

Smaller,faster,and

costlier(per byte)storage devices

Caches

Cache: The first level(s) of the memory hierarchy encountered once the address leaves the CPU.

The term is generally used whenever buffering is employed to reuse commonly occurring items such as webpage caches, file caches, & name caches.

Caching in a Memory Hierarchy

0 1 2 3

4 5 6 7

8 9 10 11

12 13 14 15

Larger, slower, cheaper storagedevice at level k+1 is partitionedinto blocks.

Data is copied betweenlevels in block-sized transfer units

8 9 14 3Smaller, faster, more expensivedevice at level k caches a subset of the blocks from level k+1

Level k:

Level k+1: 4

4

4 10

10

10

Request14

Request12

General Caching ConceptsGeneral Caching Concepts Program needs object d, which is

stored in some block b. Cache hit

Program finds b in the cache at level k. E.g., block 14.

Cache miss b is not at level k, so level k

cache must fetch it from level k+1. E.g., block 12.

If level k cache is full, then some current block must be replaced (evicted). Which one is the “victim”? We’ll see later.

9 3

0 1 2 3

4 5 6 7

8 9 10 11

12 13 14 15

Level k:

Level k+1:

1414

12

14

4*

4*12

12

0 1 2 3

Request12

4*4*12

Cache Placement

There are 3 categories of cache organization:

1. Direct-mapped.2. Fully-associative.3. Set-associative.

Direct-Mapped

The block can appear in 1 place only. Fastest & simplest organization but

highest miss rate due to contention. Mapping is usually: Block address %

Number of blocks in cache.

Data

Valid

Tag Index Offset

=

V T D

Fully-Associative

The block can appear anywhere in the cache. Slowest organization but lowest miss rate.

Tag Offset

=

V T D

Valid

V T D

…

V T D

=

=

Data

Set-Associative

The block can appear anywhere within a single set. (n-way set associative)

The set number is usually: Block address % Number of sets in the cache.

Tag Index Offset

=

V T D

Data

Valid

V T D

=

Cache Replacement

In a direct-mapped cache, only 1 block is checked for a hit, & only that block can be replaced.

For set-associative or fully-associative caches, the evicted block is chosen using three strategies:

Random. LRU. FIFO.

Cache Replacement

As the associativity increases => LRU harder & more expensive to implement => LRU is approximated.

LRU & random perform almost equally for larger caches. But LRU outperforms others for small caches.

Write Policies

1. Write Back: the information is only written to the block in the cache.

1. Write Through: the information is written to both the block in the cache & to the block in lower levels.

Reducing the Miss Rate

1. Larger Block Sizes & Caches.2. Higher Associativity.

Application Memory Management

Allocation: to allocate portions of memory to programs at their request.

Recycling: freeing it for reuse when no longer needed.

Memory Management

In many embedded systems, the kernel and application programs execute in the same space i.e., there is no memory protection.

The embedded operating systems thus make large effort to reduce its memory occupation size.

Memory Management

An RTOS uses small memory size by including only the necessary functionality for an application.

We have two kinds of memory management:

Static Dynamic

Static memory management

provides tasks with temporary data space.

The system’s free memory is divided into a pool of fixed sized memory blocks.

When a task finishes using a memory block it must return it to the pool.

Another way is to provide temporary space for tasks is via priorities:

A high priority pool : is sized to have the worst-case memory demand of the system

A low priority pool : is given the remaining free memory.

Static memory management

Dynamic memory management

employs memory swapping, overlays, multiprogramming with a fixed number of tasks (MFT), multiprogramming with a variable number of tasks (MVT) and demand paging.

Overlays allow programs larger than the available memory to be executed by partitioning the code and swapping them from disk to memory.

MFT: a fixed number of equalized code parts are in memory at the same time.

MVT: is like MFT except that the size of the partition depends on the needs of the program.

Demand paging : have fixed-size pages that reside in non-contiguous memory, unlike those in MFT and MVT

Dynamic memory management

Memory Allocation

is the process of assigning blocks of memory on request .

Memory for user processes is divided into multiple partitions of varying sizes.

Hole : is a block of available memory.

Static memory allocation

means that all memory is allocated to each process or thread when the system starts up. In this case, you never have to ask for memory while a process is being executed. This is very costly.

The advantage of this in embedded systems is that the whole issue of memory-related bugs-due to leaks, failures, and dangling pointers-simply does not exist .

Dynamic Storage-Allocation

How to satisfy a request of size n from a list of free holes. This means that during runtime, a process is asking the system for a memory block of a certain size to hold a certain data structure.

Some RTOSs support a timeout function on a memory request. You ask the OS for memory within a prescribed time limit.

Dynamic Storage-Allocation Schemes

First-fit: Allocate the first hole that is big enough, so it is fast

Best-fit: Allocate the smallest hole that is big enough; must search entire list, unless ordered by size.

Buddy: it divides memory into partitions to try to satisfy a memory request as suitably as possible.

Buddy memory allocation

allocates memory in powers of 2 it only allocates blocks of certain

sizes has many free lists, one for each

permitted size

How buddy works? If memory is to be allocated 1-Look for a memory slot of a suitable size (the

minimal 2k block that is larger then the requested memory) If it is found, it is allocated to the program If not, it tries to make a suitable memory slot. The

system does so by trying the following: Split a free memory slot larger than the requested

memory size into half If the lower limit is reached, then allocate that amount of

memory Go back to step 1 (look for a memory slot of a suitable

size) Repeat this process until a suitable memory slot is found

How buddy works?

If memory is to be freed 1. Free the block of memory 2. Look at the neighboring block - is it

free too? 3. If it is, combine the two, and go back

to step 2 and repeat this process until either the upper limit is reached (all memory is freed), or until a non-free neighbor block is encountered

64K 64K 64K 64K 64K 64K 64K 64K 64K 64K 64K 64K 64K 64K 64K64K

t=0 1024K

t=1A-64K

64K 128K 256K 512K

t=2A-64K

64K B-128K 256K 512K

t=3A-64K

C-64K

B-128K 256K 512K

t=4A-64K

C-64K

B-128K D-128K 128K 512K

t=5A-64K

64K B-128K D-128K 128K 512K

t=6 128K B-128K D-128K 128K 512K

t=7 256K D-128K 128K 512K

t=8 1024K

Example: buddy system

The problem of fragmentation

neither first fit nor best fit is clearly better that the other in terms of storage utilization, but first fit is generally faster.

All the previous schemes has external fragmentation.

the buddy memory system has little external fragmentation.

External Fragmentation–total memory space exists to satisfy a request, but it is not contiguous.

Internal Fragmentation–allocated memory may be slightly larger than requested memory; this size difference is memory internal to a partition, but not being used.

Fragmentation

Example: Internal Fragmentation

Memory Protection

it may not be acceptable for a hardware failure to corrupt data in memory. So, use of a hardware protection mechanism is recommended.

This hardware protection mechanism can be found in the processor or MMU.

MMUs also enable address translation, which is not needed in RT because we use cross-compilers that generate PIC code (Position Independent Code).

Hardware Memory Protection

Recycling techniques

There are many ways for automatic memory managers to determine what memory is no longer required

garbage collection relies on determining which blocks are not pointed to by any program variables .

Recycling techniques

Tracing collectors : Automatic memory managers that follow pointers to determine which blocks of memory are reachable from program variables.

Reference counts : is a count of how many references (that is, pointers) there are to a particular memory block from other blocks .

Example : Tracing collectors

Mark-sweep collection:

Phase1: all blocks that can be reached by the program are marked.

Phase2: the collector sweeps all allocated memory, searching for blocks that have not been marked. If it finds any, it returns them to the allocator for reuse.

Mark-sweep collection

The drawbacks : It must scan the entire

memory in use before any memory can be freed.

It must run to completion or, if interrupted, start again from scratch.

Example : Reference counts Simple reference counting : a reference count is kept for each object. The count is incremented for each new

reference, decremented if a reference is overwritten, or if the referring object is recycled.

If a reference count falls to zero, then the object is no longer required and can be recycled.

it is hard to implement efficiently because of the cost of updating the counts.

References:

http://www.memorymanagement.org/articles/recycle.html

http://www.dedicated-systems.com http://www.Wikipedia.org http://www.cs.utexas.edu http://www.netrino.com S. Baskiyar,Ph.D. and N.Meghanathan,A

Survey of Contemporary Real-time Operating Systems.