Chapter 6

Memory
 Chapter 6 Objectives

• Master the concepts of hierarchical memory

organization.
• Understand how each level of memory contributes
to system performance, and how the performance
is measured.
• Master the concepts behind cache memory, virtual
memory, memory segmentation, paging and
address translation.

2
 6.1 Introduction
• Memory lies at the heart of the stored-program
computer.
• In previous chapters, we studied the
components from which memory is built and the
ways in which memory is accessed by various
ISAs.
• In this chapter, we focus on memory
organization. A clear understanding of these
ideas is essential for the analysis of system
performance.

3
 6.2 Types of Memory
• There are two kinds of main memory: random
access memory, RAM, and read-only-memory,
ROM.
• There are two types of RAM, dynamic RAM (DRAM)
and static RAM (SRAM).
• DRAM consists of capacitors that slowly leak their
charge over time. Thus, they must be refreshed
every few milliseconds to prevent data loss.
• DRAM is “cheap” memory owing to its simple
design.

4
 6.2 Types of Memory

• SRAM consists of circuits similar to the D flip-flop

that we studied in Chapter 3.
• SRAM is very fast memory and it doesn’t need to
be refreshed like DRAM does. It is used to build
cache memory, which we will discuss in detail later.
• ROM also does not need to be refreshed, either. In
fact, it needs very little charge to retain its memory.
• ROM is used to store permanent, or semi-
permanent data that persists even while the system
is turned off.

5
 6.3 The Memory Hierarchy
• Generally speaking, faster memory is more
expensive than slower memory.
• To provide the best performance at the lowest cost,
memory is organized in a hierarchical fashion.
• Small, fast storage elements are kept in the CPU,
larger, slower main memory is accessed through
the data bus.
• Larger, (almost) permanent storage in the form of
disk and tape drives is still further from the CPU.

6
 6.3 The Memory Hierarchy
• This storage organization can be thought of as a pyramid:

7
 6.3 The Memory Hierarchy
• We are most interested in the memory hierarchy
that involves registers, cache, main memory, and
virtual memory.
• Registers are storage locations available on the
processor itself.
• Virtual memory is typically implemented using a
hard drive; it extends the address space from
RAM to the hard drive.
• Virtual memory provides more space: Cache
memory provides speed.

8
 6.3 The Memory Hierarchy
• To access a particular piece of data, the CPU first
sends a request to its nearest memory, usually
cache.
• If the data is not in cache, then main memory is
queried. If the data is not in main memory, then
the request goes to disk.
• Once the data is located, then the data, and a
number of its nearby data elements are fetched
into cache memory.

9
 6.3 The Memory Hierarchy
• This leads us to some definitions.
– A hit is when data is found at a given memory level.
– A miss is when it is not found.
– The hit rate is the percentage of time data is found at a given
memory level.
– The miss rate is the percentage of time it is not.
– Miss rate = 1 - hit rate.
– The hit time is the time required to access data at a given
memory level.
– The miss penalty is the time required to process a miss,
including the time that it takes to replace a block of memory
plus the time it takes to deliver the data to the processor.

10
 6.3 The Memory Hierarchy
• An entire blocks of data is copied after a hit
because the principle of locality tells us that once a
byte is accessed, it is likely that a nearby data
element will be needed soon.
• There are three forms of locality:
– Temporal locality- Recently-accessed data elements tend
to be accessed again.
– Spatial locality - Accesses tend to cluster.
– Sequential locality - Instructions tend to be accessed
sequentially.

11
 6.4 Cache Memory
• The purpose of cache memory is to speed up
accesses by storing recently used data closer to the
CPU, instead of storing it in main memory.
• Although cache is much smaller than main memory,
its access time is a fraction of that of main memory.
• Unlike main memory, which is accessed by address,
cache is typically accessed by content; hence, it is
often called content addressable memory.
• Because of this, a single large cache memory isn’t
always desirable-- it takes longer to search.

12
 6.4 Cache Memory
• The “content” that is addressed in content addressable
cache memory is a subset of the bits of a main memory
address called a field.
– Many blocks of main memory map to a single block of
cache. A tag field in the cache block distinguishes one
cached memory block from another.
– A valid bit indicates whether the cache block is being used.
– An offset field points to the desired data in the block.

13
 6.4 Cache Memory
• The simplest cache mapping scheme is
direct mapped cache.
• In a direct mapped cache consisting of N
blocks of cache, block X of main memory
maps to cache block Y = X mod N.
• Thus, if we have 10 blocks of cache, block 7
of cache may hold blocks 7, 17, 27, 37, . . .
of main memory.

The next slide illustrates this mapping.

14
 6.4 Cache Memory

• With direct
mapped cache
consisting of N
blocks of cache,
block X of main
memory maps to
cache block Y =
X mod N.

15
 6.4 Cache Memory
• EXAMPLE 6.1 Consider a word-addressable main
memory consisting of four blocks, and a cache with
two blocks, where each block is 4 words.
• This means Block 0 and 2 of main memory map to
Block 0 of cache, and Blocks 1 and 3 of main
memory map to Block 1 of cache.
• Using the tag, block, and offset fields, we can see
how main memory maps to cache as follows.

16
 6.4 Cache Memory
• EXAMPLE 6.1 Consider a word-addressable main
memory consisting of four blocks, and a cache with two
blocks, where each block is 4 words.
– First, we need to determine the address format for mapping.
Each block is 4 words, so the offset field must contain 2 bits;
there are 2 blocks in cache, so the block field must contain 1 bit;
this leaves 1 bit for the tag (as a main memory address has 4 bits
because there are a total of 24=16 words).

17
 6.4 Cache Memory
• EXAMPLE 6.1 Cont'd
– Suppose we need to access a
main memory address 316 (0011
in binary). If we partition 0011
b
using the address format from
Figure a, we get Figure b.
– Thus, the main memory address
0011 maps to cache block 0.
– Figure c shows this mapping,
along with the tag that is also
stored with the data.
c
The next slide illustrates
another mapping.

18
 6.4 Cache Memory

19
 6.4 Cache Memory
• EXAMPLE 6.2 Assume a byte-addressable memory
consists of 214 bytes, cache has 16 blocks, and each
block has 8 bytes.
– The number of memory blocks are:
– Each main memory address requires14 bits. Of this 14-bit address
field, the rightmost 3 bits reflect the offset field
– We need 4 bits to select a specific block in cache, so the block
field consists of the middle 4 bits.
– The remaining 7 bits make up the tag field.

20
 6.4 Cache Memory

• In summary, direct mapped cache maps main

memory blocks in a modular fashion to cache
blocks. The mapping depends on:
• The number of bits in the main memory address
(how many addresses exist in main memory)
• The number of blocks are in cache (which
determines the size of the block field)
• How many addresses (either bytes or words) are
in a block (which determines the size of the
offset field)

21
 6.4 Cache Memory
• Suppose instead of placing memory blocks in
specific cache locations based on memory
address, we could allow a block to go anywhere
in cache.
• In this way, cache would have to fill up before
any blocks are evicted.
• This is how fully associative cache works.
• A memory address is partitioned into only two
fields: the tag and the word.

22
 6.4 Cache Memory

• Suppose, as before, we have 14-bit memory

addresses and a cache with 16 blocks, each block
of size 8. The field format of a memory reference
is:

• When the cache is searched, all tags are searched

in parallel to retrieve the data quickly.
• This requires special, costly hardware.

23
 6.4 Cache Memory

• You will recall that direct mapped cache evicts a

block whenever another memory reference
needs that block.
• With fully associative cache, we have no such
mapping, thus we must devise an algorithm to
determine which block to evict from the cache.
• The block that is evicted is the victim block.
• There are a number of ways to pick a victim, we
will discuss them shortly.

24
 6.4 Cache Memory
• Set associative cache combines the ideas of direct
mapped cache and fully associative cache.
• An N-way set associative cache mapping is like
direct mapped cache in that a memory reference
maps to a particular location in cache.
• Unlike direct mapped cache, a memory reference
maps to a set of several cache blocks, similar to the
way in which fully associative cache works.
• Instead of mapping anywhere in the entire cache, a
memory reference can map only to the subset of
cache slots.

25
 6.4 Cache Memory
• The number of cache blocks per set in set associative
cache varies according to overall system design.
– For example, a 2-way set associative
cache can be conceptualized as shown in
the schematic below.
– Each set contains two different memory
blocks.

Logical view Linear view

26
 6.4 Cache Memory

• In set associative cache mapping, a memory

reference is divided into three fields: tag, set,
and offset.
• As with direct-mapped cache, the offset field
chooses the word within the cache block, and
the tag field uniquely identifies the memory
address.
• The set field determines the set to which the
memory block maps.

27
 6.4 Cache Memory
• With fully associative and set associative cache, a
replacement policy is invoked when it becomes
necessary to evict a block from cache.
• An optimal replacement policy would be able to look
into the future to see which blocks won’t be needed
for the longest period of time.
• Although it is impossible to implement an optimal
replacement algorithm, it is instructive to use it as a
benchmark for assessing the efficiency of any other
scheme we come up with.

28
 6.4 Cache Memory
• The replacement policy that we choose depends
upon the locality that we are trying to optimize--
usually, we are interested in temporal locality.
• A least recently used (LRU) algorithm keeps track of
the last time that a block was assessed and evicts
the block that has been unused for the longest
period of time.
• The disadvantage of this approach is its complexity:
LRU has to maintain an access history for each
block, which ultimately slows down the cache.

29
 6.4 Cache Memory
• First-in, first-out (FIFO) is a popular cache
replacement policy.
• In FIFO, the block that has been in the cache the
longest, regardless of when it was last used.
• A random replacement policy does what its name
implies: It picks a block at random and replaces it
with a new block.
• Random replacement can certainly evict a block that
will be needed often or needed soon, but it never
thrashes.

30
 6.4 Cache Memory
• The performance of hierarchical memory is
measured by its effective access time (EAT).
• EAT is a weighted average that takes into account
the hit ratio and relative access times of successive
levels of memory.
• The EAT for a two-level memory is given by:
EAT = H  AccessC + (1-H)  AccessMM.
where H is the cache hit rate and AccessC and AccessMM are
the access times for cache and main memory, respectively.

31
 6.4 Cache Memory
• For example, consider a system with a main
memory access time of 200ns supported by a
cache having a 10ns access time and a hit rate of
99%.
• Suppose access to cache and main memory
occurs concurrently. (The accesses overlap.)
• The EAT is:
0.99(10ns) + 0.01(200ns) = 9.9ns + 2ns = 11ns.

32
 6.4 Cache Memory
• For example, consider a system with a main memory
access time of 200ns supported by a cache having a
10ns access time and a hit rate of 99%.
• If the accesses do not overlap, the EAT is:

0.99(10ns) + 0.01(10ns + 200ns)

= 9.9ns + 2.01ns = 12ns.
• This equation for determining the effective access
time can be extended to any number of memory
levels, as we will see in later sections.

33
 6.4 Cache Memory
• Caching is depends upon programs exhibiting good
locality.
– Some object-oriented programs have poor locality
owing to their complex, dynamic structures.
– Arrays stored in column-major rather than row-major
order can be problematic for certain cache
organizations.
• With poor locality, caching can actually cause
performance degradation rather than performance
improvement.

34
 6.4 Cache Memory
• Cache replacement policies must take into account
dirty blocks, those blocks that have been updated
while they were in the cache.
• Dirty blocks must be written back to memory. A
write policy determines how this will be done.
• There are two types of write policies,write through
and write back.
• Write through updates cache and main memory
simultaneously on every write.

35
 6.4 Cache Memory
• Write back (also called copyback) updates memory
only when the block is selected for replacement.
• The disadvantage of write through is that memory
must be updated with each cache write, which slows
down the access time on updates. This slowdown is
usually negligible, because the majority of accesses
tend to be reads, not writes.
• The advantage of write back is that memory traffic is
minimized, but its disadvantage is that memory does
not always agree with the value in cache, causing
problems in systems with many concurrent users.

36
 6.4 Cache Memory
• The cache we have been discussing is called a
unified or integrated cache where both instructions
and data are cached.
• Many modern systems employ separate caches for
data and instructions.
– This is called a Harvard cache.
• The separation of data from instructions provides
better locality, at the cost of greater complexity.
– Simply making the cache larger provides about the same
performance improvement without the complexity.

37
 6.4 Cache Memory

• Cache performance can also be improved by

adding a small associative cache to hold blocks
that have been evicted recently.
– This is called a victim cache.
• A trace cache is a variant of an instruction
cache that holds decoded instructions for
program branches, giving the illusion that
noncontiguous instructions are really
contiguous.

38
 6.4 Cache Memory
• Most of today’s small systems employ multilevel
cache hierarchies.
• The levels of cache form their own small memory
hierarchy.
• Level1 cache (8KB to 64KB) is situated on the
processor itself.
– Access time is typically about 4ns.
• Level 2 cache (64KB to 2MB) may be on the
motherboard, or on an expansion card.
– Access time is usually around 15 - 20ns.

39
 6.4 Cache Memory
• In systems that employ three levels of cache, the
Level 2 cache is placed on the same die as the CPU
(reducing access time to about 10ns)
• Accordingly, the Level 3 cache (2MB to 256MB)
refers to cache that is situated between the
processor and main memory.
• Once the number of cache levels is determined, the
next thing to consider is whether data (or
instructions) can exist in more than one cache level.

40
 6.4 Cache Memory

• If the cache system used an inclusive cache, the

same data may be present at multiple levels of
cache.
• Strictly inclusive caches guarantee that all data in a
smaller cache also exists at the next higher level.
• Exclusive caches permit only one copy of the data.
• The tradeoffs in choosing one over the other
involve weighing the variables of access time,
memory size, and circuit complexity.

41
 6.5 Virtual Memory
• Cache memory enhances performance by providing
faster memory access speed.
• Virtual memory enhances performance by providing
greater memory capacity, without the expense of
adding main memory.
• Instead, a portion of a disk drive serves as an
extension of main memory.
• If a system uses paging, virtual memory partitions
main memory into individually managed page frames,
that are written (or paged) to disk when they are not
immediately needed.

42
 6.5 Virtual Memory
• A physical address is the actual memory address of
physical memory.
• Programs create virtual addresses that are mapped
to physical addresses by the memory manager.
• Page faults occur when a logical address requires
that a page be brought in from disk.
• Memory fragmentation occurs when the paging
process results in the creation of small, unusable
clusters of memory addresses.

43
 6.5 Virtual Memory
• Main memory and virtual memory are divided into
equal sized pages.
• The entire address space required by a process
need not be in memory at once. Some parts can be
on disk, while others are in main memory.
• Further, the pages allocated to a process do not
need to be stored contiguously-- either on disk or in
memory.
• In this way, only the needed pages are in memory
at any time, the unnecessary pages are in slower
disk storage.

44
 6.5 Virtual Memory
• Information concerning the location of each page,
whether on disk or in memory, is maintained in a data
structure called a page table (shown below).
• There is one page table for each active process.

45
 6.5 Virtual Memory
• When a process generates a virtual address, the
operating system translates it into a physical
memory address.
• To accomplish this, the virtual address is divided
into two fields: A page field, and an offset field.
• The page field determines the page location of the
address, and the offset indicates the location of the
address within the page.
• The logical page number is translated into a
physical page frame through a lookup in the page
table.

46
 6.5 Virtual Memory
• If the valid bit is zero in the page table entry for the
logical address, this means that the page is not in
memory and must be fetched from disk.
– This is a page fault.
– If necessary, a page is evicted from memory and is replaced
by the page retrieved from disk, and the valid bit is set to 1.
• If the valid bit is 1, the virtual page number is
replaced by the physical frame number.
• The data is then accessed by adding the offset to the
physical frame number.

47
 6.5 Virtual Memory
• As an example, suppose a system has a virtual address
space of 8K and a physical address space of 4K, and the
system uses byte addressing.
– We have 213/210 = 23 virtual pages.
• A virtual address has 13 bits (8K = 2 13) with 3 bits for the page
field and 10 for the offset, because the page size is 1024.
• A physical memory address requires 12 bits, the first two bits
for the page frame and the trailing 10 bits the offset.

48
 6.5 Virtual Memory
• Suppose we have the page table shown below.
• What happens when CPU generates address 5459 10
= 10101010100112 = 1553 16?

49
 6.5 Virtual Memory
• We said earlier that effective access time (EAT) takes
all levels of memory into consideration.
• Thus, virtual memory is also a factor in the
calculation, and we also have to consider page table
access time.
• Suppose a main memory access takes 200ns, the
page fault rate is 1%, and it takes 10ms to load a
page from disk. We have:
EAT = 0.99(200ns + 200ns) 0.01(10ms) = 100, 396ns.

50
 6.5 Virtual Memory
• Even if we had no page faults, the EAT would be
400ns because memory is always read twice: First to
access the page table, and second to load the page
from memory.
• Because page tables are read constantly, it makes
sense to keep them in a special cache called a
translation look-aside buffer (TLB).
• TLBs are a special associative cache that stores the
mapping of virtual pages to physical pages.

51
 6.5 Virtual Memory
• Another approach to virtual memory is the use of
segmentation.
• Instead of dividing memory into equal-sized pages,
virtual address space is divided into variable-length
segments, often under the control of the programmer.
• A segment is located through its entry in a segment
table, which contains the segment’s memory location
and a bounds limit that indicates its size.
• After a page fault, the operating system searches for
a location in memory large enough to hold the
segment that is retrieved from disk.

52
 6.5 Virtual Memory
• Both paging and segmentation can cause
fragmentation.
• Paging is subject to internal fragmentation because a
process may not need the entire range of addresses
contained within the page. Thus, there may be many
pages containing unused fragments of memory.
• Segmentation is subject to external fragmentation,
which occurs when contiguous chunks of memory
become broken up as segments are allocated and
deallocated over time.

53
 6.5 Virtual Memory
• Large page tables are cumbersome and slow, but with
its uniform memory mapping, page operations are
fast. Segmentation allows fast access to the segment
table, but segment loading is labor-intensive.
• Paging and segmentation can be combined to take
advantage of the best features of both by assigning
fixed-size pages within variable-sized segments.
• Each segment has a page table. This means that a
memory address will have three fields, one for the
segment, another for the page, and a third for the
offset.

54
 6.6 A Real-World Example
• The Pentium architecture supports both paging and
segmentation, and they can be used in various
combinations including unpaged unsegmented,
segmented unpaged, and unsegmented paged.
• The processor supports two levels of cache (L1 and
L2), both having a block size of 32 bytes.
• The L1 cache is next to the processor, and the L2
cache sits between the processor and memory.
• The L1 cache is in two parts: and instruction cache (I-
cache) and a data cache (D-cache).

The next slide shows this organization schematically.

55
 6.6 A Real-World Example

56
 End of Chapter 6

57