Operating System - Windows, MacOS, Linux, etc.

Most operating systems are composed of millions of lines of code.

How can the essence of each of these operating systems be captured in only one page of code?

One Page

Keeping Details

We want to keep all of the details.

But, details kill.

Q: How do we keep the details while minimizing the complexity?

(You already know the answer :-).

Decomposing Complexity

We must decompose complexity.

We must decompose complexity into layers.

Using hierarchy allows us to show details in layers.

Using components allows us to show details in hierarchical layers.

The concept of components are already be familiar to most people.

The only new idea here is the obersavation of psychology and how it affects our designs.

If we use a PL[^pl] that gives us only one layer, then we will tend to design everything in one layer.

Most PLs allow hierarchical decomposition, but do not encourage it.

Components

What is a Component?

A Component is a black box.

We can characterize components as below:

  • input commands from parent
  • input information from children
  • filter information and Send () filtered information to Parent

The points of Components are:

  • no inter-component dependencies
  • reducing information flow, instead of using an all-in-one mind-set.

Well-designed Components use no side-channels.

Well-designed Components, make all dependencies explicit (e.g. Parent/Children relationships).

Concurrency

Components are easier to design and use in the concurrent paradigm.

The synchronous paradigm makes it harder to think of components and to use them. We resort to complicated structures, like code libraries, dependency managers

Asynchrony is easy - if you don’t try to express it in the synchronous paradigm.

Multitasking, as we know it, is accidental complexity created by expressing asynchrony in a synchronous paradigm and on synchronous hardware1.

Synchronous design is a good paradigm for designing calculators.

Synchronous design is not a good paradigm for designing distributed systems2.

Dependencies

We want software components that have no dependencies between them.

We can’t scale systems that have inter-component depedencies.

Q: How can we build components that have no dependencies?

Docker

Docker is a symptom of systems that have dependencies.

We use docker instead of fixing the problem - the Elephant in the Room.

Make

Make allows us to cope with dependencies.

We use make instead of fixing the problem - the Elephant in the Room.

Package Managers

Dependency Managers, like npm, allow us to cope with dependencies.

We use dependency managers instead of fixing the problem - the Elephant in the Room.

Can Types be Componentized?

Q: Can Types be Componentized?

A: Yes.

For example:

bit -> byte -> word -> ... -> mid-level abstractions -> high level abstractions

See type stacks.

Q: How many levels of abstraction are there?

A: It depends.

We should build layers of abtraction that make sense for a given solution. Note the use of the word solution, and the avoidance on using the word PL.

Hierarchical Types

Decompose Types as Components 
bytes vs. high-level constructs ### Hierarchical Functions	 ### Hierarchical Variable Names ## Features ## File System ## Drivers ## Multitasking ### Time-Sharing ### Full Preemption ### Memory Sharing

See Also

References
Table of Contents

  1. CPUs were OK until they added The Stack (a global variable). IBM 360s did not have Stacks built into the hardware - programmers had to use BALR instructions if they really, really wanted control-flow stacks. BALR did not constrain programmers to using a global variable, programmers could build stacks in the heap, if they wanted to. Today, all stacks are in shared memory and we have to resort to accidental complexities like multitasking and full preemption. 

  2. Currently, we resort to using low-level concepts, like thread libraries, to implement distributed systems as multiple envelopes containing exactly one calculator each. This is what assembler programmers did while resisting the use of HLLs.