第6章インフラストラクチャーの抽象化層：オペレーティングシステム

By managing system resources (e.g., CPU, memory, disk, and network) efficiently, the OS ensures that no resource is underutilized or overburdened. We start with this question: What level of ease or complexity is involved in developing applications with optimal efficiency on a computer? A well-performing operating system facilitates the development of efficient applications that enhance the user experience by providing faster response times. The OS also impacts user experience positively by managing resource allocation and multitasking efficiently.

At some point, you might work for an organization that is engaged in developing a new app, and the company may ask itself this question: If we were to take the time to develop all the additional software required for our application to boot and run on raw hardware, how much faster would it be? This is the same as asking what the penalty or cost is of developing what the OS provides (i.e., sharing of the hardware among apps, and limited damage when programs have defects). The answer to this question relates to runtime efficiency. With respect to coding time efficiency, the OS includes various abstractions, interface and libraries that application programmers can use to ease the burden of software development. Having to write these from scratch without the support of an operating system would greatly increase the required coding effort and extend the development time associated with application development.

In operating systems, a policy is a way to choose which activities to perform (i.e., what needs to be done) and a mechanism is an activity that enforces policies (i.e., how to do it), which often depend on the hardware on which the operating system runs. If a process is granted resources using the first come, first served policy, then that policy may be implemented using a queue of requests. To understand how policies and mechanisms relate, consider this analogy. A car is a mechanism because it enables operations such as going, stopping, and turning that enable a driver to get from point A to point B. Notice that the mechanism (in this case, the car) does not say anything about how to use the mechanism to get anywhere in particular. The driver provides a policy by deciding on a route. Now compare a car with a public bus. In this case, the bus driver determines the policy. The same goes for the self-driving cars that are being developed these days. They serve as the mechanism as there are no decisions about routes being built into cars. Instead, a self-driving car uses machine learning to determine a policy, which includes deciding on a route, selecting the procedure on how to drive, and allowing the driver to override automated driving at any time.

The ability of a system or server to run different types of applications used by multiple users at a time on the same computer is called virtualization. Server virtualization places a software layer called a hypervisor (i.e., virtual machine monitor, or VMM) between a machine’s (i.e., server’s) hardware and the operating systems that run on it. The hypervisor creates and manages virtual machines. A virtual machine (VM) is software that is created to run like a physical computer and that operates its operating system and applications like separate physical servers. OS-level virtualization is a basic form of server virtualization. When using OS-level virtualization, there is no need for a hypervisor as the server’s OS handles all resources.

Cloud computing leverages server virtualization and makes it possible for users to commission virtual machines. In this case, the servers reside in distributed data centers provided by cloud vendors. Commissioning a virtual machine is equivalent to gaining access to a virtual computer that runs a chosen operating system while leasing the underlying hardware and only paying per use (i.e., cost of CPU, storage, and network usage).

As you’ve now learned, the OS is a system software program whose job it is to manage all the programs running on a computer. OSs enable users to run multiple applications at the same time and keep them from interfering with or crashing each other. OSs provide convenient abstractions to handle diverse hardware. They coordinate resources and protect users from each other using a few critical hardware mechanisms. OSs make it easier for developers to create applications by offering built-in features that help with managing errors. These features include fault containment, which prevents errors in one part of an application from affecting the whole system; fault tolerance, which allows the application to keep running even when errors occur; and fault recovery, which helps the system to fix itself or revert to a previous state after an error. The fact that these standard services are provided by the OS means that developers do not have to build these error management features from scratch, which simplifies the development process. Figure 6.3 shows the support layers within the UNIX/Linux system structure. The OS navigates between three layers: the user mode, where the application resides; the kernel mode, which is in effect sandwiched between the user and hardware; and the hardware, which refers to resources such as the CPU and memory. In the next sections, we will describe these modes and the features they support in detail.