Skip to main content

Computer Resources

Every digital computer system has finite resources, managed by the operating system.

One or more of these three broad categories of resources constrain (limit) the system's ability to execute each running program:

  • Memory - The availability of adequate memory, and the performance of that memory.

  • CPU - The ability of the CPU to execute the instructions contained in the program at a particular speed.

  • Input/Output (I/O) - This is a very broad category that includes storage, networking, printers, keyboard and mouse, and so forth.

Processes​

On a multitasking computer system, a process is a single instance of an executing computer program. For example, ssh is one program. If there are no running instances of ssh on a computer system, then there are no ssh processes. If one user starts one ssh session, then one there is one ssh process. If five users start a combined total of twelve ssh sessions, then there are twelve ssh processes.

Each process is assigned a Process ID (PID) and is allocated its own resources, such as memory. Processes are scheduled to run (execute) on a CPU core for a fraction of a second, and then the operating system may intervene and switch the CPU core to run another processes. However, if a process is waiting for an I/O resources, such as data from a storage device, a packet of data from the network, or a keystroke from the user, it may inform the operating system that it does not need to be executed until that resource is available. A process which is ineligible for execution because it is waiting for a resources is called a sleeping process.

Resource Conflicts​

When the resource requirements of running computer programs significantly exceed available resources, system performance will become unsatisfactory.

It is important to be able to identify the resource consumption of each process, and each operating system provides multiple tools to do this. They also provide tools to terminate processes if they are unstable or consuming excessive resources.

Challenges with Monitoring Resource Usage​

It is difficult to accurately monitor a process' resource utilization because it will change from microsecond to microsecond. In addition, some resources are very difficult to accurately measure:

Memory​

On contemporary computer systems, memory is very actively and dynamically managed by the Operating System using the hardware's virtual memory features. This enables many useful techniques, including:

  • Demand Loading - Only the portion of programs which are used are loaded. Code for features that are not used is not loaded into memory. For example, in a word processor (such as LibreOffice/OpenOffice or Word), features such as table of content generation, endnotes, and watermarks are not used in many documents, and the code for those features will not be loaded in many cases.
  • Sharing of Memory - Shared libraries (also called Dynamically Linked Libraries) may be used by many different processes. These libraries are loaded into memory only once, using demand loading, regardless of the number of processes using each library. Similarly, a program which is running in two or more processes will be loaded into memory only once, and shared between the processes. This significantly reduces memory requirements.
  • Swap - When system memory (RAM) is approaching full utilization, the operating system may take some of the least-recently-used areas of memory and place them in storage (hard disk or solid state disk) or compress them in order to avoid running out of free memory. Areas of memory swapped out may be swapped back in if they are later required.

Due to the utilization of these techniques and others, the memory consumption of a program can be calculated in many different ways. For example:

  • You could add up the full size of the program, all of the shared libraries, and the data used by a process, but that would yield a large number (a pessimistic view of memory consumption). The process is likely not occupying that much memory, and terminating the process will typically release only a fraction of that memory because most of the shared libraries will still be loaded into memory and in use by other programs.
  • You could add up just the portion of the program in memory plus the data area in memory used by a process, but this would understate the full memory utilization because it does not include the shared libraries nor the portion of the program or data which has been swapped out or which has not (yet) been demand-loaded.

Therefore resource monitoring tools may present several different statistics about memory usage for each process, or may try to present a rough approximation of the amount of total system memory which can be reasonably attributed to a process.

CPU​

Each CPU core is independently capable of executing programs. Most modern CPUs have multiple cores (Symmetric Multi Processing, SMP), enabling multiple programs to be executed simultaneously. Many modern CPUs can execute more than one thread on a core (Symmetric Multi Threading, SMT), meaning that they have hardware support for rapid switching between two or more processes (or, in some cases, sub-processes called threads) on one core -- so a 4-core CPU may be marketed as capable of running up to 8 tasks. However, this does not mean that 8 programs can be fully executed at once!

To use an analogy, a SMP processor is a bit like an apartment building, where each core is like an apartment and can fully accommodate a family (whether a single, couple, roommates, or parents and children), or, in the case of the CPU, fully execute a process. SMT capability is a bit like Airbnb rentals, which enable those units to be shared out quickly when they're not in use by the main occupants, or in the case of the CPU, quickly run other threads when the core (or part of the core) is not in use.

Further complicating the issue is the fact that CPUs may have multiple cores of different specifications -- some fast, power-hungry cores used for heavy tasks, and other slower, energy-efficient cores for lighter tasks. Cores may be shut down and started up by the OS as needed. For example, some Seneca lab computers have an Intel CPU with 12 cores consisting of 10 performance cores (capable of running 2 threads) and 2 efficiency cores (capable of running 1 thread). In a similar way, a Google Pixel 8 phone has 4 energy-efficient/low-performance Arm Cortex-A510 cores, 4 mid-range Cortex-A715 cores, and one high-energy-usage/very high-performance Cortex-X3 core.

In addition, current CPUs can operate at a range of speeds, and the OS and hardware work together to dynamically tune the speed to balance temperature, energy usage, and performance.

Therefore, statistics reporting the percentage of CPU capability in use by a process may report the percentage of a core's capability in use, either with or without considering SMT capabilities, or they may report the percentage of the full CPU's capability (including all cores). When cores of different specifications are part of a single CPU, it may be difficult to determine the relative ratio of performance between the different types of cores, especially as the speed of those cores is being adjusted -- and therefore utilization percentages are almost always a rough approximation.