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

How long does an OS last? Did companies stop developing new OSs? How can the current OS survive? To answer these questions, we need to discuss concepts such as paging, TLBs, disks, storage latency, and multicores as well as virtual machines (VM).

Many of the basic ideas in OSs were developed 30–50 years ago, when technology was very different. The question is not only whether these ideas will still be relevant in the future, but whether they are relevant even today. After all, technology has changed considerably over the last thirty or so years. For example, CPU speeds went from 15 MHz in 1980 to 2.5+ GHz in 2024, a 167-fold increase. Memory size went from 8 MB to 16+ GB, a 2,000-fold increase. Disk capacity went from 30 MB to 2+ TB, a 6,667-fold increase. Disk transfer rate went from 2 MB/sec to 200+ MB/sec, a 100-fold increase. Network speeds went from 10 Mb/sec to 10+ Gb/sec, a 1,000-fold increase. As you can see, there were huge increases in size, speed, and other capabilities.

As you may recall, paging is a storage mechanism that allows processes to be retrieved from secondary memory and moved to main memory using pages. In the 1960s, paging originally touted disk speed latency of 80 ms, a data transfer rate of 250 KBs/sec, memory size of 256 Kbytes. Thus, for 64 pages, it took 6.4 sec to replace all of the memory to address individual page faults and 1 sec to address sequential page faults. Today, we have disk speed latency of 10 ms, a data transfer rate of 150+ MB/sec, and memory size of 64+ GB. For 16,000,000+ pages, it takes 44+ hours to replace all of memory to address individual page faults, and 320+ sec to address sequential page faults. Therefore, we cannot afford to page something out unless the system is going to be idle for a long time. But the real question is: does paging make sense anymore as a mechanism for the incremental loading of processes? The answer is yes, but by reading the entire binary at once because 15 MB of binary takes 0.1 sec to read.

TLBs have not kept up with memory sizes; 64 entries provide 256 KB coverage. In the mid-1980s, this was a substantial fraction of memory (i.e., 8 Mbytes). Today, TLBs can only cover a tiny fraction of memory. Some TLBs support larger page sizes of 1 Mbyte or 1 GB, but this complicates kernel memory management.

Disk capacity has increased faster than access time; storage access latency for disks is around 10 ms, and it is around 100 µs for flash memory. There are new nonvolatile memories, such as Intel’s 3D XPoint, that improve the latency to 100ns–300ns.

Chip technology improvements allowed processor clock rates to improve rapidly. Unfortunately, however, faster clock rates mean more power dissipation, and now power limitations limit improvements in clock rate. Chip designers are now using technology to put more processors (cores) on a chip. In general, all OSs must now be multiprocessor OSs. However, it is not clear how to utilize these cores, and application developers must write parallel programs, which is very hard.

Lastly, the current/hot trend for OS development is the data center, which coordinates thousands of machines working together trying to achieve very low-latency communication.

As nearly every person and business on the planet uses computers today, their reliability and security are increasingly essential. At the same time, there is growing concern about whether the underlying technologies we are relying on to power OSs will become obsolete soon. And there are also questions about what will replace OSs. Check out the debate on what will replace OS (https://openstax.org/r/76replaceOS) and see whether you share any of the concerns.

As you learned earlier in this chapter, a virtual machine is a software emulation of a physical computer that creates an environment that can execute programs and manage operations as if it were a separate physical entity. This emulation allows multiple operating systems that are isolated from each other to run concurrently on a single physical machine. In essence, a VM provides the functionality of a physical computer, including a virtual CPU, memory, hard disk, network interface, and other devices.

Recall that the underlying technology enabling VMs is called a hypervisor or virtual machine monitor (VMM). This technology resides either directly on the hardware (Type 1 or bare-metal hypervisor) or on top of an operating system (Type 2 or hosted hypervisor). The hypervisor is responsible for allocating physical resources to each VM and ensuring that they remain isolated from each other. This isolation ensures that processes running in one VM do not interfere with those running in another and thereby enhances security and stability. VMs are widely used for a variety of purposes, including server virtualization, software testing and development, and desktop virtualization. Virtual machines have become a fundamental component of cloud computing, as they allow cloud providers to offer scalable and flexible computing resources to users on a pay-as-you-go basis.

Figure 6.38 illustrates the difference between a Type 1 virtual machine monitor and container environment such as via Docker. A container is a standardized unit of software that logically isolates an application, enabling it to run independently of physical resources.

When the complete OS is running within a VM, then the system will be called a guest operating system. VMs are heavily used in cloud computing such as Microsoft Azure, Amazon Web Services, Google Cloud Platform, and IBM Cloud.