An operating system is a software that sits above hardware in order to make life easier for application programmers. For most Internet routers, time-critical packet forwarding runs directly on the hardware (Fig. 2.10) and is not mediated by an operating system. Less time-critical code runs on a router operating system that is stripped down, such as Cisco’s IOS. However, to improve end-to-end performance for, say, Web browsing, an implementor needs to understand the costs and benefits of operating systems.
Abstractions are idealizations or illusions we invent to deal with the perversity and irregularity of the real world. To finesse the difficulties of programming on a bare machine, operating systems offer abstractions to application programmers. Three central difficulties of dealing with raw hardware are dealing with interruptions, managing memory, and controlling I/O devices. Operating systems provide the abstractions of continuous computation, infinite memory, and straightforward I/O to address these issues. A good abstraction increases programmer productivity but has two costs. First, the mechanism implementing the abstraction has a price. For example, scheduling processes can cause overhead for a Web server. A second, less obvious cost is that the abstraction can hide power, preventing the programmer from making optimal use of resources. For example, operating system memory management may prevent the programmer of an Internet lookup algorithm from keeping the lookup data structure in memory in order to maximize performance. We now provide a model of the costs and underlying mechanisms of the process (Section 2.4.1), virtual memory (Section 2.4.2), and I/O (Section 2.4.3) abstractions. More details can be found in Tanenbaum (1992).