cs459 - Operating Systems: Introduction

advertisement
Operating Systems
• Definition: Software that creates an environment so computer can
perform useful tasks
• What is in an operating system?
– Loose definition: Everything that ships with the computer
– More precise definition: Software that runs at all times (kernel).
The rest is either a system utility or an application program
• Possible focus
–
–
–
–
–
–
Provide a good user experience
Responsive to user commands
Minimize idle time and maximize throughput
Respond to real time constraints
Maximize connectivity
Efficient sharing distributed resources
Computer System Components
1. Hardware – One or more CPUs, I/O devices and controllers, bus
or switch connections, memory
2. Operating system modules: kernel, bootstrap (start up
firmware), device drivers (for I/O), and disk resident subsystems
3. System utilities – Sysgen (System generation) software,
defragmenters, disk editors, etc.
4. Application programs- Word processors, compilers, web
browsers, database systems, video games
5. Users: People, machines, other computers
Types of Operating Systems
• Batch: Small resident kernel sequences jobs for one or more batch streams
using a job control interface. Goal: minimize setup time, minimize idle time
• Real time: Often part of special purpose embedded systems with no user
interface. Goal: satisfy well-defined hard or soft time constraints
• Time shared: Responds to multiple users at workstations. Scheduling is
based on real time clock “slices”. Goal: Responsive to users
• Desktop systems: Single users, with multitasked and multi-programming
capabilities. Goal: user friendly and responsive
• Multiprocessor: Tightly coupled processors sharing memory and
peripherals. Goal: maximize throughput and reliability (redundancy)
• Multi-computer: Loosely coupled, often heterogeneous systems working
cooperatively. Goal: efficient and secure client-server or peer-to-peer intercomputer communication
• Grid-based: loosely coupled, heterogeneous, geographically separated.
Communicate through LAN or WAN; peer-to-peer or client-server. Goal:
utilized distributed resources to address grand challenge problems
• Handheld systems: Slow, limited in processor speed, available power and
user interfaces. Goal: support user-friendly portable applications, maximize
wireless connectivity, optimize battery life, optimize execution efficiency
Real Time System Constraints
• Hard (System fails if constraints not satisfied)
– Guarantee that real time constraints complete on-time.
– Not supportable by general purpose operating systems.
– Data is stored in ROM or in short term memory; no secondary
storage, virtual memory, or advanced operating system features.
– Examples: Respond to sensor control inputs, assist scientific
experiments, weapon systems, medical imaging
• Soft (Best effort made to meet constraints)
– Priority-based, delays are bounded.
– Not useful for hard restraints like when a robotic arm movement
cannot be stopped on time or if a gyroscope falls without fast
enough CPU signals supplied.
– Examples: streaming of audio and video, web cams, stock
market activity
Parallel Processing
Peer to Peer or Client Server
• Multiprocessors (tightly coupled)
– Can be Less expensive because they share peripherals
– Can be more reliable:
• Service continues upon processor failure (graceful degradation)
• One processor picks up functionality upon failure (fault tolerant)
– Operation can be
• Asymmetric: Each processor performs a separate task
• Symmetric: All processors perform the same tasks simultaneously
• Clustered systems (loosely coupled via local LAN)
– High availability service
– Asymmetric: one machine is in hot-standby mode
– Symmetric: multiple host running and monitoring each other
• Grids: loosely connected and geographically separated
Computing System Software
Computing System Hardware
Dual-Mode Control
Enables the Operating system to protect itself from users and executing software
• Mode bit: Provided by hardware for dual-mode operation
– User mode: Executing privileged instructions will automatically trap to
the operating system
– kernel mode: Full access to the system is enabled
• Processing
– At boot time: System is automatically in kernel mode
– During System call: Mode automatically changes kernel mode
– After responding to system call: Restore application environment, set
mode to user, and reschedule the application
– Illegal operations: A trap (exception) occurs. Control immediately
transfers to the operating system, which creates a data dump and
terminates the executing program
• Note: Early PC's had no mode bit; applications could (and did) often
crash the system
System Timer
Enables the Operating system to maintain ultimate control over the system
• Definition: A device that signals at regular intervals (i.e. every
millisecond)
• Processing: A timer signal occurs
Control transfers to the operating system in kernel mode
Operating System decrement a system counter
IF counter = 0
IF executing application exceeds run time limits (i.e. infinite loop)
Create core dump and terminate the application
ELSE IF process is compute bound (i.e. allocated time slice expires)
Save application environment and schedule another application
ELSE return control to the application
System Call Processing
Note: The OS must save the application environment, process
the request, and then restore the environment when the
application is rescheduled.
System Boot
1. Reset event: hardware sets the program counter to a predefined location
2. A firmware bootstrap loader executes to:
a. Execute diagnostics
b. Loads a boot block from a fixed disk location
3. The boot block then loads the entire operating system into memory
4. The Operating system initializes itself and begins to execute
Firmware Notes
• Definition: Firmware is a set of instructions programmed persistently on
a hardware device (i.e. ROM or EEPROM)
• Some embedded devices store their entire operating system in firmware
• Firmware updates: ROM-based firmware requires chip changes;
EEPROM-based firmware allows updates after special command given
System Software Components
• Sysgen:
– Purpose: install/configure an operating system onto specific hardware
– Goal: user friendly installation process
• Kernel: The part that is always in memory
– Provides a basic interface for programs and device drivers
– Include: scheduler, I/O subsystem, Memory Management module,
and many others
• User Interface: Command line interface, shell, or GUI enabling the
operating system to communicate with users
• System Programs: System level applications that normally reside
on disk. Examples include loaders, compilers, editors, disk utilities, batch
systems, communication services, system usage monitors, etc.
The Operating System
• Operating System Functions
–
–
–
–
–
–
–
–
–
–
–
Create a secure and stable user-friendly environment
Efficiently schedule and execute programs (processes ) and threads
Read and write from system hardware devices
Connect to local and wide area networks
Manage memory allocation and protect against illegal access
Define and maintain the directory and access to secondary storage
Allocate and share resources efficiently and fairly
Maintain system security
Enforce scheduling priorities
Recovery from system faults
Perform low-level system services for applications
• Implementation
– Higher level language: Portable, fast development, more maintainable
– Assembly language: fast, less portable, difficult to maintain
– Typical operating systems: 95% C, 5% Assembly
Optimizing System Utilization
Note: A single user generally cannot keep the CPU and I/O devices busy
• Multiprogramming: Concurrent process execution
– A set of jobs is kept in memory (or partially on disk)
– The OS job scheduler module picks one to execute
– A OS switch occurs when it blocks (ex: for I/O)
• Multitasking/Time sharing: Multi-user responsiveness
–
–
–
–
Extends multiprogramming
OS switches between jobs frequently
User Response time should be < 1 second
Each user has at least one program executing
• Multithreading: Multiple paths of execution in a process
– Time consuming work is done in separate threads
– Threads are spawned by processes
– They have a program counter, register allocation, and stack but share
all other resources with the parent process
• Considerations
– Swapping moves processes in and out of memory to run
– Virtual memory allows execution of processes that don't fit in memory
Process Management
• A process is a program that is executing.
– Programs are passive (blocks of executable instructions/data)
– Processes are active (A program together with its active state)
• Operating System responsibilities
–
–
–
–
–
–
–
Give process the resources they need
Provide an API so processes can "talk" to the OS
Protect processes from one another
Allow processes to communicate with one another
Implement fair share scheduling algorithms
Terminate and reclaim reusable resources
Create and manage execution of threads
• Note: Typically systems have many processes running. These
include applications, system processes, and background services
Scheduling
• Short Term (Must be extremely efficient)
–
–
–
–
–
Look at queue of processes that are ready to run
Decide which has the highest priority
Restore its runtime environment
Set the system to user mode
Transfer control to the process
• Long Term (Low priority background process)
– Decide if system load is at an acceptable level
– Launch processes waiting to enter the system
Von Neumann model
Process Control Block
Definition: An object that maintains the state of a process
• A program counter and registers for all threads in a
process
• Pointers to the resources allocated to a process
– File and device handles
– Mutexes
– Connections to remote resources
• Physical and logical memory allocated to this process
(which can be contiguous or not, in memory or on disk)
• Initialization data
• Runtime statistics and accounting information
• Links to shared memory and connections to other
processes
Memory Management
• Memory: volatile large array of bytes directly accessible by
the processor. Can be DRAM (dynamic ram – one transistor
per bit) or RAM (volatile transistor based storage)
• Operating System responsibilities
– Keep track of which parts of memory are currently in use
– Transfer processes (or parts thereof) and data in and out
of memory
– Allocate and release memory space as needed
– Maintain sections of memory for caching disk operations
– Control shared memory accessible to multiple processes
– Map logical program addresses into physical memory and
maintain portions that are on disk and not contiguous
– Prevent processes from overwriting each other's memory
Protect and Manage Memory
Operating System
Operating System
P1 part 1
P1 part 2
p1 part 3
Disk
Memory Management Examples
Free BSD
MS-DOS
(a) no process (b) process running
Storage Management
• Provide a uniform logical view of data
– A logical storage view abstracts physical properties
– Each medium is controlled by a storage device
– Examples: disk, flash disks, cd, dvd, tape
• OS responsibilities
– Create the file system structure of files and directories
– Mount and dismount devices
– Maintain secondary storage by effectively mapping files
onto available areas of the system mediums
– Backup files onto stable storage media
– Establish transparent connections to remote data
– Establish access control mechanisms
– Schedule writes and reads for maximum efficiency
Storage Hierarchy
• Storage systems are organized in hierarchy
– Higher levels are faster, smaller, and are more likely to be volatile
– Lower levels are slower, larger, and are less likely to be volatile
• Cache: Temporarily copy data into faster storage
–
–
–
–
Locality principle: Data accessed clusters in local areas
Load policy: Retrieve the data needed plus surrounding data
Replacement policy: Choice of cache items to expel when the cache fills
Processing:
• The system looks to a faster cache before accessing the slower hierarchy levels
• Higher levels in the storage hierarchy can serve as caches for the lower levels
• On updates, lower levels are either immediately updated (write through) or
updated only when an item is expelled (delayed write) from the cache
• Cache coherence (consistency of multiple caches) is an important consideration
in parallel systems which contain multiple caches
• Caching can be explicitly OS controlled or implicitly done in hardware
Secondary Storage Hierarchy
Non
Volatile
Volatile
Slow Cheap
Fast Expensive
Caching: Temporarily storing recently accessed data higher in the hierarchy
Storage Hierarchy Performance
Cache Coherence
• We must always use the most recent value,
no matter where it is in storage hierarchy
• A Multiprocessor environment must provide
cache coherency. All CPUs have the most
recent value in their local cache
• Distributed environments are even more
complex
I/O Management
• Purpose: hide hardware specifics from users
• I/O subsystem responsible for
–
–
–
–
–
Buffering (storing data temporarily while it is being transferred)
Caching (storing parts of data in faster storage for performance)
Spooling (writing to disk instead of directly to the device)
Uniform device-driver API to abstract device specifics
Managing drivers (software modules) for specific hardware devices
• Device Controllers
–
–
–
–
–
Operate simultaneously and autonomously with CPU
Are limited instruction set processors to solely manage devices
Are responsible for a particular type of device
Contain local memory buffers and hardware registers to control I/O
Example: SCSI (small computer-systems interface) handles seven
or more devices
The Interrupt
Definition: A hardware signal that
forces a program counter (PC) change
• Interrupts occur when
– Hardware devices complete operations
– Software requests a service or attempts an illegal operation (a trap)
• Processing (Interrupts give operating systems ultimate control)
– Interrupt architecture saves the interrupted instruction address
– Incoming interrupts are disabled to avoid lost interrupts.
– Control may transfer control via a
• Low memory device indexed interrupt vector (array) of addresses
• A fixed location that polls to identify the type of interrupt
– The CPU state is saved (PC and register contents).
– Separate routines process the interrupt
– The CPU state is restored and execution continues
Approaches for Handling Devices
1. Device driver API: a uniform mechanism to abstract hardware
operations from the rest of the system
2. Programmed I/O: Computer starts an operation and then waits in a
tight loop for the operation to complete
3. Polling: A loop that inquires and responds to each device
4. Interrupts (A special I/O signal raised by a device)
a.Computer starts an operation
b.Computer does other things
c. Computer transfers data, pice by piece, when interrupts occur
5. Direct Memory Access (DMA)
a.Computer starts an operation
b.Computer does other things
c. Device transfers a block using main memory by cycle stealing
d.Computer responds to interrupt when device i/o is done
How do Interrupts Work?
• Device raises an interrupt signal
• Interrupt signal interrupts whatever the computer
was doing, lower priority interrupts get disabled
• The address of the executing instruction, registers,
etc get saved (normally with a hardware stack)
• An interrupt vector (array) contains addresses of
device drivers
• Device driver processes the interrupt
• Incoming lower level interrupts are re-enabled
• Control goes back to the OS short term scheduler
Trap: Interrupts generated by software
Device Controllers
Slow Devices
Interrupt-driven IO
Fast Devices
Direct Memory Access,
possible cycle stealing
Bus versus Switched
Architecture
Switched Architecture Advantage: Avoids cycle stealing
Interrupt Timeline
Synchronous & Asynchronous I/O
Synchronous
Asynchronous
Device-Status Table
Protection and Security
• Protection: Internal mechanism to control user and process access
to system resources
– Specifies and implements the controls available
– Distinguish between authorized and unauthorized use
– Detect and recover from hardware faults
– Examples: infinite loop, malicious processes, disk corruption
• Security: Internal or external attack defense mechanisms
– Operating system middleware level approaches
– Examples: denial-of-service, worms, viruses, identity theft
• Techniques to enforce protections and security
– Distinguish between groups of users: user ID and group ID
– Privilege escalation: process request additional rights (setuid call)
– Access Control: rwx rights (chmod)
Network Services
• Communications – Exchange of Information
– To processes in the same computer
– To processes on remote network computers
• Protocol Stack Layers (ISO)
1. Physical layer – Standardized electrical interfaces (RS 232C)
2. Data Link Layer – Send, receive, detect errors, retransmit (Ethernet)
3. Network Layer – Routing (IP)
4. Transport Layer – connection-oriented (TCP) or connectionless
(UDP) packet handling
5. Session Layer – Synchronization for checkpoints
6. Presentation Layer – Format of the message fields
7. Application Layer – FTP, HTTP, e-mail, etc.
Note: Layers 5 and 6 are often not used
Note: Middleware consists of background service
processes, which provide communication and data
management facilities to application programs
User Interface
Definition: The mode of communication between the operating system and users
• Embedded systems: special purpose buttons and displays
• Unix and batch systems: command line interface (CLI)
– Direct command entry
– Fetch command and execute
– Fast commands execute directly, others launch system programs
• Windows and IOS: Graphics User Interface (GUI)
–
–
–
–
Point and click: mouse, keyboard, and monitor
Touch: finger movements trigger actions
Icon based: files, programs, actions, etc show as icons
Various types of mouse clicks respond accordingly
• Hybrids: Both CLI and GUI components
–
–
–
–
Microsoft Windows is GUI with CLI “command” shell
Apple Mac OS X is GUI interface with various UNIX shells
Solaris is CLI with optional GUIs (Java Desktop, KDE)
Handheld has terminal apps with touch controls
Fault Recovery and Detection
• OS constantly checks for possible errors
–
–
–
–
–
Faulty disk sectors (maintain spare tracks and sectors)
Physical memory failure (skip use of memory areas)
Program faults (rollback, checkpoint, restart)
Network connection problems (retransmission)
Device redundancy (switch devices automatically upon failure)
• Debugging aids:
–
–
–
–
Provide logs and dumps along with the ability to analyze problems
Single step debugging capabilities
Program trace facility
Execution histograms
• Backup and clean up facilities
– Defragmenter, disk verifier, firewall
System Call – OS Application API
• The OS implements a table of available system calls, each with a welldefined interface and supported by compiler run-time libraries
• User program invoke calls to run-time libraries, which execute
privileged instructions and traps to the OS kernel. The OS processes
system call requests, returning status and return values
Standard API Example
Java read() method
byte[] b - the buffer to store read data
int off - the starting offset in b
int len - the maximum number of bytes=
System Call: Copy File Example
• Low level assembly: Execute
designated privileged instruction,
which traps to the OS
• Low level C: Utilize libraries to
execute privileged instructions
• Higher level programmers:
Use class-based interfaces
• Cross-platform libraries
– Win32 API: windows
– POSIX API: POSIXcompliant systems
– Java API: Java JVM
Copy File application system call log
System Call arguments
1. First approach: Pass in hardware registers
Disadvantage: There may not be enough registers
2. Second approach: Pass in an register addressed array.
Advantage: There is no argument limit
Note: Linux and Solaris use this approach
3. Push parameters onto a special stack
Advantage: most flexible
Disadvantage: loss of efficiency
Parameter Passing via Table
Operating System Design Goals
Tradeoffs: flexibility, efficiency, kernel size, maintainability
• Maximum functionality/efficiency using minimal memory
– Example: MSDOS, which is monolithic, non-modular, unprotected
• Maintainability (Use a higher level language (i.e. C) where possible)
– Layered design: Lower levels request services from higher levels
• Advantage: Easy testing and replacing layer implementations
• Less efficient: multiple levels of system calls from layer to layer
• Difficulty: no clear cut way to assign system functions to layers
– Modular: Separate object oriented modules with well-defined interfaces
– Micro kernel
• Perform as many functions as possible in user mode
• Communication using a message passing paradigm
• Easier to extend and port; is more reliable and secure.
• Degraded performance; inefficient user to kernel communication
OS Organization Examples
Solaris Kernel Modules
Layered Approach
Unix Kernel
•
•
•
•
•
Each layer built on top of lower layers.
The bottom layer (layer 0), is hardware
The highest (layer N) is user interface.
layers use services of lower-level layers
Easy to debug, and test replacement
layers
OS Policies and Mechanisms
Policy (Design Goal)
Mechanism (Implementation)
Program Independence
Restrict memory access
Fair scheduling of resources
Priority scheduling
Support many devices
Standard device driver interface
Transparent input and output
Privileged instructions
Plug and play
Check for new hardware
Fault recovery
Device redundancy
Authentication/authorization
Access control lists, passwords
Run very large programs
Virtual memory
Handle critical sections
Semaphores
Concurrency
Multithreading, Multiprogramming
Virtual Machines
• Virtual machines provide
– API identical to bare hardware
– Allocate of portions of disk to
each virtual machine
– Spooling to virtual print devices
– Workstation virtual console
– Combine software emulation or
hardware virtualization
• Advantages:
– Operating systems research
– Cross-platform testing
• Disadvantages:
– No direct sharing of resources
between virtual machines
– Significant loss of performance
– No sharing of resources
between the virtual machines
The operating system illusion of
multiple processes executing on
their own bare-hardware processor
VMware Architecture
Java Virtual
Machine
• Java uses architecture
neutral byte codes that
are executed by a Java
Virtual Machine (JVM)
• The JVM consists of a
Class loader, Class
verifier, Runtime
interpreter
• The just-in-time
compilation results in
performance that is often
close to native-code
execution
The JX Operating Systems
1. Each domain in JX is a separate JVM, with its own heap and garbage collection
2. Domain 0 is the micro kerrnel, written in C and assembly, and providing low level
hardware access. Other domains are written entirely in Java.
3. Cross communication implemented through portals
4. The language semantics eliminates the need for using memory management
facilities; however, virtual memory then is not possible
Download