CIS 307: Views, Components, and Architectures of Operating Systems

[Introduction], [Resource Manager], [Virtual Machine], [Reactive System], [Components], [Monolithic Architecture], [Layered Architecture], [Virtual Machine Architecture], [Micro-Kernel Architecture] [Multi-Threaded Kernel]

Introduction

Here is a possible definition of an Operating System (paraphrased from Brinch Hansen):
An Operating System is a set of manual and automatic procedures that enable a group of people to use a computer installation efficiently.
The word Efficiently is very ambiguous. What are the Objective Functions to be used to evaluate efficiency? Here are some possibilities:

Functionality
Performance: Time (Response Time=Turnaround Time, Latency), Utilization (CPU, Memory, I/O, Communication Channels, ..)
Convenience

Of course these objective functions may not all agree. For example, cost and functionality may conflict, as well as response time and utilization.

Resource Manager

We can think of an operating system as a Resource Manager. Here are some of the resources managed by the OS:

Processors, Main memory, Secondary Memory, Devices, Information.

While managing these resources the OS is concerned with general issues like:

Efficiency, Protection, Security, Privacy, Reliability.

Virtual Machine

We can think of an operating system as a Virtual Machine standing between our programs and the bare hardware, thus providing a more convenient, extended machine.

or alternatively

Reactive System

We can think of an operating system as a Reactive System answering requests coming from user programs (supervisory calls+exceptions) and from devices (interrupts).

Components

Here are the basic components of an OS:

Process Management: Create/Delete/Suspend/Resume Processes, Inter Process Communication (IPC)
Memory Management: Allocate and Deallocate Physical Memory
Virtual Memory Management: Maintain mapping from virtual to physical memory, decide on size of memory to allocate to processes, enforce replacement policy
I/O Management: definition of Device Drivers providing a uniform interface to devices
Secondary Storage Management: allocation and deallocation of blocks, recovery block management.
File Management: naming, directories, creation, use, and deletion of files, locks, protection, backups, accounting.
Network Services: support for various protocols and communication facilities.
Command Language Interpreter(CLI): interface to operator and users that gives them the ability to give commands to the OS.
Information Services: Show/Set commands
Accounting Services: Quotas, usages, and Costs.

Some functionality, such as Protection, Security, Privacy, Reliability are realized across many components.

Monolithic Architecture

The oldest architecture used for operating systems is the Monolithic Kernel

A monolithic kernel can be easy to implement if we are ready to accept very high latency [i.e. we service requests withour worrying about the urgency of other requests], otherwise it may be with a structure that is very difficult to understand. Monolithic kernels are an architecture whose time has come and gone.

Layered Architecture

Dijkstra introduced the Layered Architecture for operating systems when he developed in the 60's the THE operating system.

It is not always easy to have a truly layered architecture since some functionalities are naturally described as mutually dependent. In addition a layered architecture is often inefficient since it requires a high number of traversals of interfaces.

Virtual Machine Architecture (VM370)

This architecture creates a virtual 370 machine for each "user" and the user can choose which OS to run on that virtual machine.

The virtal machine architecture is elegant, but it does not deal with the questions of resource management, or of responsiveness.

Micro-Kernel Architecture

A very modern architecture (Mach, OSF, and NT) is the Micro-Kernel Architecture.

The micro-kernel architecture strives to take out of the kernel as much functionality as possible, so as to limit the code executed in privileged mode and to allow easy modifications and extensions. The micro-kernel architecture allows us to build a number of different operating systems all using the same micro-kernel, Each operating system will make use of different system processes. Notice that little is in the kernel. But what is there has great importance. Interactions between processes involve kernel intervention. Thus the efficiency of the transfer from user to kernel mode and back must be very high.

Multi-Threaded kernel

Many believe that the micro-kernel architecture is inherently inefficient and that the use of a more extensive but pre-emptible kernel is preferable. The idea here is to do things almost as in the monolithic kernel, but now allowing concurrency, and preemption, within the kernel itself. The argument for this change is made on the basis of efficiency. All the context switches required in the micro-kernel architecture to go from kernel to system tasks and back are replaced by subroutine calls within the kernel itself, with great saving in time. Of course, writing the kernel becomes substantially more difficult because now within a single address space and mode we can have more than one concurrent action. Sun Solaris, for example, uses this architecture.
[Notice that if we think in terms of preemption (control is taken away suddenly form an executing activity), we see that the multithreaded kernel provides preemption both within processes and the kernel. Traditional kernel provide provide preemption, if at all, only at the user level.]

http://www.cis.temple.edu/~ingargio/cis307/readings/intro.html

ingargiola.cis.temple.edu