Transcript Interface - Duke University

CPS110 / EE 153:
Intro to Operating Systems
Jeff Chase
August 25, 2008
About the guy I got these slides
from (Landon Cox)
 Background
 BS Math/CS: Duke, ’99
 PhD EECS: Michigan, ’05
 Research interests
 OS, p2p, economics, security, mobility
 Why am I a professor?
 Research and teaching are a lot of fun
 Explaining things improves my understanding
About me (Jeff Chase)
 Background
 BS Math/CS: Dartmouth, back in the 1980s sometime
 PhD CS: University of Washington (Seattle), ’95
 Research interests
 OS, networked systems, Internet service infrastructure,
utility computing, energy/green, cool new stuff
 Why am I a professor?
 Research and teaching are a lot of fun, etc.
 Explaining things improves my understanding
 Office with view of tower
Syllabus: prerequisites (CPS)
 CPS 100
 Basic data structures
 Allocating memory on the stack versus from the heap
 CPS 104





Basic computer architecture, ISAs
Registers: stack pointer, PC, general-purpose
Virtual memory translation
Page tables
TLB, caching
Syllabus: lectures and textbook
 Lecture notes on the web (125 pages)
 Exams based on content of lectures
 Textbooks




Not required
On-line: Saltzer and Kaashoek
“Modern Operating Systems” is OK
Useful: Storage, data, and information systems
($15 on Amazon)
Syllabus: discussion sections
 Two sections, starting next week
 MW 2:50- 4:05
 F 2:50 – 4:05 (sometimes)
 Teaching Assistant
 Amre Shakimov ([email protected])
 Seasoned and energetic
 Undergraduate Teaching Assistant
 Matt Jacobson
Syllabus: projects
 Where you will learn the most
 4 projects




0: very simple intro to C++
1: building a user-level thread package
2: building a virtual memory manager
3: hack into a vulnerable system
 Projects aren’t long, but are difficult
 Only 100-1,000 lines/code, but many hours
 Everything is in C++
 Project 0 has been posted today
Syllabus: homework problems
 Posted on web by Friday
 Should be done before discussion section
 Not graded, but count toward participation
Syllabus: project groups
 All projects done in groups of 2 or 3
 Email groups to [email protected]
 By Friday (August 29)!
 Group members will rate each other
 Procedure for firing, quitting in syllabus
Syllabus: project auto-grading
 All projects are auto-graded
 Allows groups to get immediate feedback
 Use submit110 script on cs machines
 One submission/group/day gets feedback
 Can’t use to debug your project
 Any group member’s submission counts
More on the auto-grader
 Very limited feedback: correct or incorrect
 Doesn’t say what is wrong
 Still have to write a test suite (except P0)
 Don’t rely on auto-grader feedback alone
 To get more useful feedback
 Come talk to us!
 We will provide many office hours every week
 (double office hours week before a deadline)
Syllabus: project timelines
 Due at 6pm, accepted until 11:59:59pm
 Auto-grader clock is the one that counts
 Last submission to auto-grader is final
 3 late days/group/semester
 Intended for unexpected problems
 No extensions
 Start early!
Syllabus: project collaboration
 Ok, among groups
 C++ syntax, course concepts
 “What does this part of the handout mean?”
 Not ok, among groups
 Design/writing of another’s program
 Includes prior class solutions
 “How do I do this part of the handout?”
 We use automated similarity-detection software
 Just changing the variable names won’t save you
 If in doubt, ask me
Syllabus: grades, exams
 Projects: 35%
 Midterm: 30%
 mid-October
 Final: 30%
 December
 Participation: 5%
Projects and exams
 The two are not independent
 Familiarity with projects is critical to doing well on exams
 I like to ask questions about projects on exams
 “Extend Project X to include this functionality”
 Know your project!
 You can assign roles to different people
 But each member must understand all aspects
Syllabus: environment
 Linux/GNU environment
 Need to sign-up for term CS account
 Use the form on the cs.duke.edu/csl page
 Send CS login name to [email protected]
 Can login into linux.cs.duke.edu
 Use this account for all auto-grading
Syllabus: getting help
 Newsgroup
 http://courses.duke.edu
 Office hours
 With me:
 With Amre:
 Don’t email Amre or me directly
 Post to the newsgroup, which we monitor
Grades from last semester
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
1
2
3
4
5
6
Some kind of A
7
8
9
10 11 12 13 14 15 16
Some kind of B
sko D
Questions about the mechanics?
Goals for CPS 110
 First part: demystify the operating system
 How does my computer start running?
 How does a program load into memory?
 Second part: demystify Internet systems
 How does my email know where to go?
 Why is Google so fast?
 How is everything “virtualized”?
Duke course view of computers
CPS 1,6,100,108
What’s
CPSmissing?
110
CPS 104
Applications
Ideas 
high-level programming languages
compiling, reading programs off disk,
getting program into memory, reading
keyboard, starting the computer, saving
files, filenames, networking
Hardware
Assembly language program 
gates
Thinking about interfaces
 Consider the Java language and its key word “interface”
 What is a Java object?
 List of methods and collection of internal state
 What is a Java interface?
 Set of methods associated with an object that a programmer can call
 What do those methods do?
 Invoke code (let the object do work on the caller’s behalf)
 Mutate the object’s public/private state
 Why are interfaces useful?
 They provide an “abstraction” or simplification
 Callers don’t have to know an object exact type
OS terminology

Key terms: interface, resource (cpu, mem, etc), abstraction, virtual
 Define an interface in terms of resources
 An interface is a set of primitives or operations
 Interfaces provides access to resources
 What do we mean by abstraction?
 How resources are presented to a client
 Can think of as an illusion that makes resources easier to program
 What does it mean to virtualize something?
 Provides an abstraction (simple way to manipulate resources)
 (mostly) disallow direct access to reality/resources
What is an operating system?
 Program that runs on CPU, (mostly) like any other
Applications
OS
Hardware
“Virtual machine”
Interface
“Physical machine”
Interface
 Virtual interface should be simpler than physical
What is an operating system?
Applications
What interface does
the OS present?
“Virtual machine”
Interface
OS
“Physical machine”
Interface
Hardware
What interface does
the hardware present?
Instruction set:
Load/store, mem, regs
Hardware-software stack
Applications
Threads
Virtual
Memory
System
Calls
Socket
Traps
Ether,
80211
OS
Atomic
Test/Set
Hardware
Page
Tables
OS vs user-level programs
Familiar view
Alternate view
User program
OS
OS
How do programs start?
Tasks outside program? (net recv)
How to prevent CPU hogging?
User program
User program
OS runs first, calls program
Programs run until they return control to OS
(by themselves or forced by hardware)
Then OS calls another program
Key question: who calls whom?
Functions of the OS
1. Illusionist
 Makes computer seem nicer than it really is

Examples?






Programs seem to have their own CPU
AFS: single, unified file system
Name data with human-readable names
Directories
Packets get lost; OS makes net look reliable
Disk is slow; OS makes it look fast via caching
Functions of the OS
1. Illusionist
 Makes computer seem nicer than it really is
2. Government
 Divides hardware resources among competing programs

What hardware resources does the OS manage?




Processor
Memory
Network
Disk
Functions of the OS
1. Illusionist
 Makes computer seem nicer than it really is
2. Government
 Divides hardware resources among competing programs
 Taxes programs (OS needs CPU, memory to run)
 Taken for granted when it works, cursed when it breaks
Why study operating systems?
 Very few of you will ever write one …
 Illusionist, govn functions appear in many domains
 Google provides the illusion of a single web server
 Word does background spell checking
 Design principles
 Proper abstractions, caching, indirection
 Concurrency, naming, atomicity, authentication
 Protection, resource multiplexing (fairness)
 How does OS create the illusions we know/love?
Hints for designing systems
 What is a system?
 Components, interconnections
 Interfaces, environment
 Systems do something for their environs
 Exhibit this behavior via interface
 Cleanly divides the world in two
 Parts of the system + the environment
Systems from 10,000 feet
Component
Component
Component
Component
System
Environment aka “the client”
Why is designing systems hard?
1. Emergent properties
 Can’t predict all component interactions
 Millennium bridge
 Synchronized stepping leads to swaying
 Swaying leads to more forceful synchronized stepping
 Leads to more swaying …
2. Propagation of effects
3. Incommensurate scaling
4. Trade-offs
Why is designing systems hard?
1. Emergent properties
2. Propagation of effects





Want a better ride so increase the tire size
Need a larger trunk for the larger spare
Need to move the back seat forward
Need to make front seats thinner
Leads to worse driver comfort than before
3. Incommensurate scaling
4. Trade-offs
Why is designing systems hard?
1. Emergent properties
2. Propagation of effects
3. Incommensurate scaling




Consider the giant mouse
Weight ~ size3 (volume)
Bone strength ~ size2 (cross section area)
An elephant sized mouse is not sustainable
4. Trade-offs
Why is designing systems hard?
1.
2.
3.
4.
Emergent properties
Propagation of effects
Incommensurate scaling
Trade-offs




“Waterbed effect”
Push on one end, and the other goes up
Spam filters and smoke detectors
False positives vs false negatives
Why is designing systems hard?
1.
2.
3.
4.

Emergent properties
Propagation of effects
Incommensurate scaling
Trade-offs
In the immortal words of HT Kung
 “Systems hard. Must work harder.”
History of operating systems
 History dominated by two trends
 Increasingly inexpensive hardware
 Increased software complexity
 Microsoft embodies tension between these trends
 MS gained 90% market share by running on cheap hw
 Supporting all that hardware complicates the OS
 (3rd-party drivers responsible for vast majority of crashes)
 How is Apple’s strategy different?
 Jobs chooses the hardware you will run
 HW-to-app control reduces complexity, choice, discount
First phase: single operator
 One goal: make it work
 Interactive (user has entire machine to herself)
 Users sign up, get room for two hours at a time
 “OS” is really just a library compiled into program
What is wrong with this timeline?
CPU utilization is awful
Since CPUs were expensive, this mattered
Second phase: batch processing
 Goal: improve CPU, I/O utilization
 Machine is no longer interactive
 Users submit program (stack of cards) to queue
 One job at a time, CPU idle during I/O, I/O idle during CPU
 OS is a batch monitor + library of services
 Loads program, runs program, prints results
 Loads next program …
Second phase: batch processing
 Goal: improve CPU, I/O utilization
 Machine is no longer interactive
 Users submit program (stack of cards) to queue
 One job at a time, CPU idle during I/O, I/O idle during CPU
 What key OS function starts to matter now?
 Protection: programs must not corrupt monitor
 Programs must relinquish CPU to monitor
Second phase: batch processing
 Goal: improve CPU, I/O utilization
 Machine is no longer interactive
 Users submit program (stack of cards) to queue
 One job at a time, CPU idle during I/O, I/O idle during CPU
 Why wasn’t protection an issue before?
 No batch monitor to corrupt
 Person in lab coat took CPU back from program
Third phase: multi-program batch
 Goal: overlap CPU, I/O




When one job is reading from disk, run another job on CPU
Use DMA + interrupts to allow background I/O
DMA: devices write to program memory
Interrupts: devices can tell CPU the I/O is done
Job 1
Job 2
Third phase: multi-program batch
 Goal: overlap CPU, I/O
 What are the OS’s new responsibilities?
 Switch between processes
 Manage multiple I/Os across devices
 Protect processes from each other
Job 1
Job 2
Fourth phase: time-sharing
 Goal: keep efficiency, restore interactivity
 Key insight: humans are really just slow I/O devices
 Switch between programs during think-time
Job 1
Job 2
Job 3
Increased complexity:
• Many jobs
• Outstanding reqs
• Many job sources
Fifth phase: personal computing
 What are PC operating systems most like?
 As PC prices dropped, single-operator became feasible
 OS was again just a library of services (MS-DOS)
 With one user, do jobs need to time-share?
 Early PC OSes could only do one thing at a time
 Everything waited while printing/loading a program (Mac < X)
 Need protection if I’m the only one using the PC?
 Protect me from myself (or my buggy software)
 Early PCs provided no protection
 (why Windows before XP, Mac before X were awful)
 PC operating systems are basically time-sharing OSes now
Operating system complexity
 Windows XP
 > 40 million lines of code
 Most of this code is device drivers (not written by MS)
 Windows NT took 7 years to develop
 Only worked well years after it shipped
 Windows 2000
 Shipped with 63,000 “potential known defects”
 Hot research area
 Simplify, automatically find OS bugs