QMCS 360 Class Notes # 6	Dr. Rick Smith Quantitative Methods and Computer Science
Dr. Smith Home | Research | Classes | Blackboard | Cryptosmith | QMCS Home | UST A-Z | UST Home last update: Tuesday, September 27, 2005

Homework

Executing the Operating System

Options

Kernel runs as "itself" independent of process environment

Kernel runs as part of the user's process but in kernel mode

Kernel has its own processes

Running without a process

Need special mechanisms to do multi-step operations that pause and resume

Can't do file I/O directly

Need to do mode switch, but not process switch, to do OS services

Running in the user process

OS code running in user mode can do I/O subject to user's privileges

OS code needs to be visible to all processes

Need to do mode switches, but not process switches, to do OS services

Running own processes

Must do mode switch and process switch for some services

OS code can do I/O subject to process-specific privileges

OS code does not need to be in all processes address spaces

Enforces a modular design on the OS

least privilege

separation of duties

Good in a microkernel architecture - minimize amount of kernel code

Good in multiprocessor environment - distribute OS operations among processors

Threads and Processes

Distinction

A process has an entire "context" as we talked about earlier - states, queues, wait conditions, RAM allocations, device allocations, etc.

A thread has its processor context, a process state (run, wait, etc), and it may be on a queue - a "lightweight" process

What are threads for?

Foreground/background: foreground processes the GUI, background updates the window contents

Asynchronous operation - having part of the program read ahead from the hard drive so the data is already in RAM when needed

Network server - have one thread for each incoming connection

Thread operation

Typical states

Running

Ready

Blocked

Typical operations

Spawn

Block

Unblock

Finish

Thread implementation

User level threads

User program has its own dispatching software (in a library)

Scheduling and "interrupts" via "signal" operations

Library intercepts blocking I/O and substitutes for nonblocking I/O, then schedules a different thread - if the OS supports this

Allows for application-specific scheduling

Can't take advantage of multiprocessing

Kernel level threads

Kernel implements the threads - user program doesn't need its own scheduler and/or dispatcher

Kernel can schedule different threads on different processors

Blocking I/O doesn't block all threads - kernel can schedule another thread

Kernel thread switching / scheduling takes 10-30 times as long as user level switching

Combined approach

Implement user level threads and schedule them via 2 or more kernel level threads