7 minute read
Let’s first look at the hardware of a computing system. This computing system consists of:
In addition, a computing system can have higher level applications. These are the “programs” that we use every day on our computer:
The Operating System is the layer of software that sits between the hardware components and the higher level applications.
You don't want to have to worry about the nuts and bolts of interacting with storage devices when you are writing an application. The operating system provides a higher level abstraction, the file, with a number of operations - like read and write - which applications can interact with.
Operating system allocates memory for applications, schedules them for execution on the CPU, controls access to various network devices and so on.
When applications are running concurrently, the operation system has to ensure that they can do what they need to without hurting one another. For example, memory allocated to each application must not be readable/writable from another application.
An operating system is a layer of systems software that
Certain operating systems may target the desktop environment, while others may target an embedded environment, while still others may target a mainframe or a server environment.
For our purposes, we will focus mainly on operating systems for desktop environments and embedded environments.
For desktop operating systems we have:
For embedded operating systems:
To achieve its goals, an operating systems provides a number of high level abstractions, as well as a number of mechanisms to operate on these abstractions.
Examples of abstractions include:
Corresponding mechanisms could be:
Operating systems may also integrate specific policies that determine exactly how the mechanisms will be used to manage the underlying hardware.
For example, a policy could determine the maximum number of sockets that a process has access to.
Let's look at an example of memory management.
The main abstraction here is the memory page, which corresponds to some addressable region of memory of some fixed size.
The operating system integrates mechanisms to operate on that page like allocate, which actually allocates the memory in DRAM. It can also map that page into the address space of the process, so that the process can interact with the underlying DRAM.
Over time, the page may be moved to different spaces of memory, or may be moved to disk, but those who use the page abstraction don't have to worry about those details. That's the point of the abstraction.
How do we determine when to move the page from DRAM to disk? This is an example of a policy, and one such implementation of that policy would use the least-recently-used (LRU) algorithm, moving pages that have been accessed longest ago onto disk.
We want to incorporate flexible mechanisms that can support a number of policies.
For the example of memory, we can have many policies: LRU, LFU (least-frequently used), random. It is a good design strategy to create our memory management mechanisms such that they can generalize to these different policies.
In different settings, different policies make more sense.
Understanding the common case - which may change in different contexts - helps the OS implement the correct policy, which of course relies on generalized mechanisms.
Computer systems distinguish between at least two modes of execution:
Because an OS must have direct access to hardware, it must operate in kernel mode.
Applications generally operate in user-mode.
Hardware access can only be utilized in the kernel mode from the OS directly.
Crossing from user-level to kernel-level is supported by most modern operating systems.
As an example, the operating system can flip a bit in the CPU that allows applications executing instructions to have direct access to hardware resources. When the bit is not flipped, operations are forbidden.
When privileged instructions are encountered during a non-privileged execution, the application will be trapped. This means the application's execution will be interrupted, and control will be handed back to the OS.
The OS can decide whether to grant the access or potentially terminate the process.
The OS also exposes an interface of system calls, which the application can invoke, which allows privileged access of hardware resources for the application on behalf of the operating system.
Operating systems also support signals, which is a way for the operating system to send notifications to the application.
Begin within the context of a currently executing process. The process needs access to some hardware, and thus needs to make a system call. The application makes the system call (potentially passing arguments), and control is passed to the operating system, which accesses the hardware. Execution control (as well as any necessary data) is passed back from the operating system to the application process.
In terms of context switching, the process involves a change from user-mode to kernel-mode to user-mode.
Not necessarily a cheap operation to make a system call!
Arguments to system call can either be passed directly from process to operating system, or they can be passed indirectly by specifying their address (pointers?)
In synchronous mode, the process waits until the system call completes. Asynchronous modes exist also.
User/Kernel transitions are very common and useful throughout the course of application execution.
The hardware supports user/kernel transitions.
For example, the hardware will cause a trap on illegal executions that require special privilege.
Hardware initiates transfer of control from process to operating system when a trap occurs.
User/Kernel transition requires instructions to execute, which can take ~100ns on a 2Ghz Linux box.
In addition, the OS may bring some data into the hardware cache, which will bounce out some other memory used by another application. Accessing data out of cache, say from main memory, may have as much as 100x impact on the performance of some subsequent actions.
An operating system provides applications with access to the underlying hardware.
It does so by exporting a number of services, which are often directly linked to the components of the hardware:
In addition, some services are even higher level abstractions, not having a direct hardware component. For example, the filesystem as a service.
Basic services include:
Historically, operating systems have a monolithic design.
Everything is included off the bat!
This makes the operating system potentially very large.
This type of operating system has a basic set of services and APIs that come with it.
Anything not included can be added, as a module.
As a result, each application can interface with the operating systems in the ways that make most sense to it.
Dynamically install new modules in the operating system!
Microkernels only require the most basic operating system components.
Everything else (including file systems and disk drivers) will run outside of the operating system at user-level.
This setup requires lots of interprocess communication (IPC), as the traditional operating system components run within application processes.
As a result, the microkernel often supports IPC as a core abstraction.
Kernel consist of several logical components
Each subcomponent within the three above can be modified or replaced 'cause modularity!
OMSCS Notes is made with in NYC by Matt Schlenker.