Csc-246 (Ncsu) Operating Systems

Front 1

Distributed Computing Advantages

Back 1

- Resource sharing

- Computational Speed up

- Reliability: Recovery from a failed host

- Communcation: between users

- Cost of effectiveness

Front 2

Distributed Computing Sources of Unreliability

Back 2

- Routing choices

- Competing Applications

- Failed Connections

- Congestion

Front 3

Physical Layer of Network Interconnection

Back 3

A hardware device to encode/decode a bit stream from the medium. Hardware manufacturers will need to agree upon standards.

Front 4

Data link Layer of Network Interconnection

Back 4

- Organize data into fixed- or variable-length frames

- Correct errors introduced by physical layer

- A header to help with error correction

- No routing decisions (LAN)

Front 5

Network Layer of Network Interconnection

Back 5

-Organize data into variable-length packets

- Additional header info to make routing data for sending data across multiple connections

- Example: IP

Front 6

Transport Layer of Network Interconnection

Back 6

- Provide useful semantics on top of network layer

- Create illusion of messages or connections for clients

- break messages into packets and re-assemble for on arrival

- Examples: TCP and UDP

Front 7

Internet Protocol (IP)

Back 7

- best-effort packet delivery service between hosts

- Each host represented by a 32-bit IP address

- No error detection

- No congestion control

Front 8

User Datagram Protocol (UDP)

Back 8

- best effort message delivery service between applications-

- No need to establish connection first

- Error detection within message

- port numbers to dispatch messages to the right process

Front 9

Transmission Control Protocl (TCP)

Back 9

- Connection must be established first

- In-order byte stream delivery

- packet boundaries hidden

-message boundaries not respected

- Congestion control by sender if network is overwhelmed

Front 10

Internet Heirarchy (top to bottom)

Back 10

- Application layer

- transport layer

- network layer

- data link layer

-physical layer

Front 11

The Sockets API

Back 11

- A collection of system calls to create and use TCP connections (and other protocols)

Front 12

Sockets in Java

Back 12

java.net-ServerSocket

java.net.Socket

java.net.DatagramSocket

Server:

- ss = new ServerSocket(PORT);

- Socket sock = ss.accept();

Client:

- Socket sock = new Socket(HOST, PORT);

Front 13

Remote Procedure Call(RPC)

Back 13

- A framework to simplify application distribution by breaking an application at procedure call boundaries

Front 14

RPC Semantics

Back 14

- Exactly once Semantics

- Every call is carried out exactly one

- At-least-once semantics

- Client keeps retransmitting until it receives desired response

- At-most-once semantics

- Every call is carried out at most once

Front 15

Network Operating System

Back 15

User/ Programmer must be aware of the most boundaries

- Collection of libraries/tools to access remote resources

Front 16

Distributed Operating system

Back 16

- User/programmer doesn't need to constantly think about host boundaries

- Remote and local resources look and behave the same

Front 17

Host-based processing

Back 17

- Server does almost all the work

Front 18

Server-based processing

Back 18

- Server does all application processing

- Client just does user interface

Front 19

Client-based processing

Back 19

- Client does all application processing

- Server just does validation and execution of requests

Front 20

Cooperative processing

Back 20

- Split application logic and computational load between client and server

Front 21

Protection: Object

Back 21

A thing you may want to use

Front 22

Protection: Operation

Back 22

Something you can do with an object

Front 23

Protection: Access right

Back 23

A rule that says it's OK to do a particular operation on a particular object

Front 24

Protection: Protection Domain

Back 24

Our name for the thing that has access rights

Front 25

Principle of Least Privilege

Back 25

Programs should have permissions to do only what they needed to do

Front 26

Access Matrix: Owner Rights

Back 26

Who can add/remove access rights for an object

Front 27

Access Matrix: Switch Rights

Back 27

To change domains

Front 28

Access Matrix: Control Rights

Back 28

To decide who can switch to a given domain

Front 29

Access List

Back 29

Store permissions within the object

Front 30

Capability List

Back 30

Store permissions with the domain

Front 31

Lock and Key Representation

Back 31

List of keys for each domain

List of locks for each object

Perform an operation if there's a match

Front 32

Types of Security Violations

Back 32

- Breach of Confidentiality

- Breach of Integrity

- Breach of Availability

- Theft of Service

-Denial Of Service

Front 33

Types of Attacks

Back 33

- Back-door

-Trojan Horse

- Stack/Buffer overflow

- Logic Bomb

-Timing Attack

-Virus

Front 34

Symmetric Encryption

Back 34

- One key to encrypt and ecrypt (DES, RC5, AES)

Front 35

Asymmetric Encryption

Back 35

Two keys, one to encrypt and one to decrypt

Front 36

Cryptographic Hash

Back 36

Maps a message to small hash value

- Given a hash value, it's hard to find any message that hashes to it

- Example: MD5, SHA-1

Front 37

Seek Time

Back 37

Time to move head to desired cylinder

- About 10-15 ms

Front 38

Rotational Latency

Back 38

-Time for the desired sector to rotate under the head

- About 4.15 ms for 7200rpm

Front 39

Transfer Rate

Back 39

Speed for reading/writing disk

- 30 MB/s - 100 MB/s

Front 40

SCAN Disk scheduling

Back 40

Move disk arm back and forth, service requests as you go

Front 41

C-SCAN disk Scheduling

Back 41

Similar to SCAN, but runs all the way back to the other edge without servicing requests

Front 42

LOOK and C-LOOK

Back 42

Stop at the last (innermost or outermost) request. In the case of C-LOOK go back to the cylinder of the first request.

Front 43

Data Striping

Back 43

How we interleave data across disks

Fine grained:

- bit-interleaved

Coarse grained:

-block-interleaved

Front 44

Host Attached Storage

Back 44

- Via a device controller, and a physical connection to that controller

- Typically using specialized hardware to communicate (SATA) with device

Front 45

Network Attached Storage

Back 45

- Pack up messages destined for the disk

- Wrap them in another protocol and send to another host

- unpack them there, and execute them

- Uses general-purpose networking

- Uses standard remote file system protocls (NFS)

Front 46

Storage Area Network (SAN)

Back 46

- Uses special-purpose interconnections (fibre channel)

- Encapsulates disk protocol (SCSI over fibre channel)

Front 47

CPU vs. GPU

Back 47

CPU good for running few threads, each doing it's own thing.

GPU good for running thousands of threads, each doing the same thing.

Front 48

CUDA Blocks

Back 48

A block of threads

Front 49

CUDA grid

Back 49

A collection of blocks, where each thread can figure out what it's supposed to do

Front 50

CUDA Kernel

Back 50

A name for a block of code to be run in parallel

(__global__void myFunction). Call by specifying:

myFunction<<blocksPerGrid, threadPerBlock>>()

Front 51

Thrashing

Back 51

The system isn't getting much work done because it's spending all its time servicing page faults

Front 52

Translation Look-aside Buffer

Back 52

A special cache for each CPU core

Maps page number straight to frame number

Front 53

The Memory-Management Unit

Back 53

A hardware device that handles mapping logical to physical address

Front 54

Page Table Base Register

Back 54

Tells hardware where the current page table is located

Front 55

Necessary Conditions for Deadlock

Back 55

- Mutual Exclusion

- Hold and Wait

- No Preemption

- Circular Wait

Front 56

Deadlock Prevention

Back 56

Build a system where deadlock can't occur

- Prevent ONE necessary condition

Front 57

Deadlock Detection

Back 57

Let deadlocks happen, notice when they occur, and respond somehow

Front 58

Deadlock Avoidance

Back 58

For each request, decide whether the process should wait. Use Banker's Algorithm

Front 59

How to read and Write Binary data to File

Back 59

fread(&a, sizeof(int), l, infile);

fwrite(&a, sizeof(int), l, outfile);

Front 60

How to control for Byte Order

Back 60

int a;

a = htonl(a);

or if receiving, convert to host's order:

a = ntohl(a);

Front 61

Requirements for Critical-Section Solution

Back 61

- Mutual Exclusion

- Progress

- Bounded Waiting

Front 62

Semaphore PsuedoCode

Back 62

- sem s = 1

- acquire(s)

- release(s)

Front 63

POSIX Semaphore Library

Back 63

sem_init(sem_t* sem, int value);

sem_Destroy(sem_t* sem);

sem_wait(sem_t* sem);

sem_post(sem_t* sem);

Front 64

Java Semaphore Library

Back 64

- new Semaphore(5);

- s.acquire()

s.release()

s.acquireUninterruptibly()

Front 65

Monitor psuedo code

Back 65

condition c;

c.wait()

c.signal()

Front 66

POSIX API for Monitors

Back 66

pthread_mutex_t lock;

pthread_mutex_init(&lock, NULL);

pthread_mutex_lock(&lock);

pthread_mutex_unlock(&lock);

pthread_cond_t cond;

pthread_cond_Wait(&cond, &lock);

pthread_cond_signal(&cond);

pthread_cond_broadcast(&cond);

Front 67

Java Monitors

Back 67

Just use synchronized methods and blocks

- any object is a condition variable.

- use wait() and notify() or notifyAll()