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 the end of this section, you will be able to:

• Discuss key concepts related to memory

• Evaluate dynamic storage management solutions

• Discuss the differences between virtual and physical memory

As you have learned, memory plays a huge role in OSs. Here, we discuss the memory multiplexing, linkers and dynamic linking, dynamic storage management, virtual memory, and demand paging.

Different processes and threads share the same hardware. It is therefore necessary to multiplex the CPU (i.e., temporal execution), memory (spatial access), and disks and devices. As discussed earlier, the complete working state of processes and/or kernels is defined by its data (i.e., memory, registers, and disk). For the sake of safety, security, and reliability, processes should be barred from having access to each other’s memory. Dividing the capacity of the communication channel into multiple logical channels is considered memory multiplexing. There are several concepts that are critical to memory multiplexing, namely, isolation, sharing, virtualization, and utilization.

As you learned earlier in this chapter, isolation is important because it ensures that the multiple programs that are running concurrently on the same CPU and memory operate independently without interfering with each other’s execution or data. In memory multiplexing, isolation is achieved through a set of technologies that prevent distinct process states from colliding in physical memory due to unintended overlap (i.e., overlap control). These technologies aim to, for example, prevent process P1 from spying on process P2. Or, if process P1 has a bug, they ensure that this bug does not impact process P2. There are many isolation mechanisms, including:

• User/kernel mode flag is a register that represents the CPU mode as user mode or kernel mode. As we have learned, the CPU boots in kernel mode, then it marks the flag as kernel mode. When the user starts any application, the CPU marks the flag as user mode.

• Address space boundaries protect the kernel and address space programs from each other.

• System call interface is the programming interface for application users to process a system call. As shown in Figure 6.25, a system call is executed by the user mode to request the kernel mode to perform a specific action (e.g., syscall () function).

Time slicing provides a time frame for each process to run in a preemptive multitasking CPU such that each process runs every single time slice. If the process finishes the job before the time slice, it releases the CPU and does not need to be swapped out. If the time slice ends and the process did not finish the job, the CPU shifts it out to the end of the processes queue. For example, assume we have three processes P1 with execution time 3 ms, P2 with execution time 4 ms, and P3 with execution time 2 ms, and a time slice of 2 ms. Figure 6.26 illustrates how the CPU manages the processing using time slice and indicates in which time slice each process completes execution.

When multiple processes can use the same piece of data concurrently, it is called sharing. The option for overlapping processes should be available when desired for efficiency and communication reasons. Memory sharing improves the performance of the system because the data is not copied from one address space to another, so memory allocation is done only once.

With respect to memory, virtualization is a technique that gives an application the impression that it has its own logical memory and that it is independent from the available physical memory. Thus, in virtualization, there is a need to create the illusion that there are more resources than those that actually exist in the underlying physical system. There are two approaches of memory virtualization: full virtualization and guest modification. When multiple operating systems run concurrently on a single physical machine, fully isolated from each other, by emulating hardware resources through a hypervisor, it is called full virtualization. For full virtualizations, all OSs expect contiguous physical memory that starts from physical address 0. In the context of virtualization, guest modification refers to altering the guest operating system or its configuration to improve compatibility, performance, or integration with the virtualization environment or hypervisor. Guest modification modifies the OS to avoid using instructions that virtualize inefficiently. An optimal use of limited resources is warranted to guarantee a high level of utilization.

Processes use different amounts of memory, and their memory needs change over time. Whenever a new process cannot fit into contiguous space in physical memory, it results in fragmentation (specifically, external fragmentation). When the memory blocks cannot be allocated to the processes due to their small size and the blocks remain unused, this problem is called fragmentation. There are two types of fragmentation: internal fragmentation and external fragmentation. When the process is allocated a block and its size exceeds the process size, it leaves part of the memory allocated unused and results in internal fragmentation. In the external fragmentation, the total space that is needed for the process is available, but we can’t use it because the space is not contiguous.