From the Operating Systems course (CMPSCI 377) at UMass Amherst, Fall 2007.

1. Operating Systems
CMPSCI 377
Virtual Memory
Emery Berger
University of Massachusetts Amherst
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science

2. Virtual Memory
Virtual, not physical memory

Divided into pages

Page tables, TLB

Provides isolation, protection, optimization

Managed using eviction policies

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 2

3. Demand-Paged Virtual Memory
Key idea: use RAM as cache for disk

OS transparently moves pages

Requires locality:

Working set (memory referenced recently)

must fit in RAM
If not: thrashing (nothing but disk traffic)

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 3

4. Virtual Memory Pages
Memory divided into

fixed-sized pages (e.g.,
4K, 8K)
Allocates pages to frames

A
in memory
OS manages pages

Moves, removes,

reallocates
Copied to and from disk

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 4

5. Demand-Paging Diagram
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 5

6. Paging + Locality
Most programs obey

90/10 “rule” A
90% of time spent

accessing 10% of A
B
memory
B
Exploit this rule:

Only keep “live” parts

of process in memory
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 6

7. Virtual Memory
Processes use virtual addresses

Addresses start at 0

OS lays process down on pages

MMU (memory-management unit):

Hardware support for paging

Translates virtual to physical addresses

Uses page table to keep track of frame

assigned to memory page
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 7

8. Mapping Virtual to Physical
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 8

9. Address Translation
Powers of 2:

Virtual address space:

size 2m
Page size 2n

High-order m-n bits

of virtual address
select page
Low order n bits

select offset in page
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 9

10. Paging Hardware: Diagram
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 10

11. Translation Lookaside Buffer (TLB)
TLB: fast, fully associative memory

Caches page table entries

Stores page numbers (key) and frame (value)

in which they are stored
Assumption: locality of reference

Locality in memory accesses =

locality in address translation
TLB sizes: 8 to 2048 entries

Powers of 2 simplifies translation

of virtual to physical addresses
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 11

12. TLB Action
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 12

13. TLB: Diagram
v = valid bit: entry is up-to-date

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 13

14. Cost of Using TLB
Measure in terms of memory access cost

What is cost if:

Page table is in memory?

Page table managed with TLB?

Large TLB:

Improves hit ratio

Decreases average memory cost

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 14

15. Sharing
Paging allows sharing of memory across

processes
Different virtual addresses,

point to same physical address
Shared stuff includes code, data

Compiler marks “text” segment (i.e., code) of

applications (e.g., emacs) - read-only
OS: keeps track of such segments

Reuses if another instance of app arrives

Copy-on-write

Can greatly reduce memory requirements

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 15

16. Key Policy Decisions
Two key questions: (for any cache):

When do we read page from disk?

When do we write page to disk?

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 16

17. Reading Pages
Read on-demand:

OS loads page on its first reference

May force an eviction of page in RAM

Pause while loading page = page fault

Can also perform pre-paging:

OS guesses which page will next be needed,

and begins loading it
Most systems just do demand paging

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 17

18. Demand Paging
On every reference, check if page is in

memory (valid bit in page table)
If not: trap to OS

OS checks address validity, and

Selects victim page to be replaced

Invalidates old page

Begins loading new page from disk

Switches to other process

(paging = implicit I/O)
Note: must restart instruction later

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 18

19. Swap Space
Swap space = where victim pages go

Partition or special file reserved on disk

Sometimes: special filesystem (“tmpfs”)

What kind of victim pages can be evicted

without going to swap?
Size of reserved swap space limits what?

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 19

20. Swap Space
Swap space = where victim pages go

Partition or special file reserved on disk

Sometimes: special filesystem (“tmpfs”)

What kind of victim pages can be evicted

without going to swap?
Read-only (“text”), untouched anonymous pages

Size of reserved swap space limits what?

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 20

21. Swap Space
Swap space = where victim pages go

Partition or special file reserved on disk

Sometimes: special filesystem (“tmpfs”)

Only read-only or undirtied victim pages can

be evicted without going to swap
“text” (code), untouched anonymous pages

Size of reserved swap space limits

total virtual memory
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 21

22. Cost of Paging
Worst-case analysis – useless

Easy to construct adversary example:

every page requires page fault
A, B, C, D, E, F, G, H, I, J, A...
size of available memory
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 22

23. Page Replacement Algorithms
MIN, OPT (optimal)

RANDOM

evict random page

FIFO (first-in, first-out)

give every page equal residency

LRU (least-recently used)

MRU (most-recently used)

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 23

24. MIN/OPT
Invented by Belady (“MIN”)

Now known as “OPT”: optimal page

replacement
Evict page to be accessed furthest in the

future
Provably optimal policy

Just one small problem...

requires predicting the future
Still useful point of comparison

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 24

25. MIN/OPT example
Page faults: 5

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 25

26. RANDOM
Evict any page

Works surprisingly well – why?

Theoretically: very good,

but not used in practice:
takes no advantage of locality
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 26

27. LRU
Evict page that has not been used in

longest time (least-recently used)
Approximation of MIN if recent past is good

predictor of future
Variant of LRU used in all real operating

systems
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 27

28. LRU example
Page faults: ?

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 28

29. LRU example
Page faults: 5

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 29

30. LRU, example II
Page faults: ?

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 30

31. LRU, example II
Page faults: 12!

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 31

32. FIFO
First-in, first-out: evict oldest page

Also has competitive ratio k

But: performs miserably in practice!

LRU takes advantage of locality

FIFO does not

Suffers from Belady’s anomaly:

More memory can mean more paging!

LRU & other “stack” algs. do not

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 32

33. FIFO & Belady’s Anomaly
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 33

34. MRU
Evict most-recently used page

Shines for LRU’s worst-case: loop that

exceeds RAM size
A, B, C, D, A, B, C, D, ...
size of available memory
What we really want: adaptive algorithms

(e.g., EELRU – Kaplan & Smaragdakis)
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 34

35. The End
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 35

36. LRU: No Belady’s Anomaly
Why no anomaly for LRU?

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 36