Fundamentals of Operating Systems

Most operating systems are general purpose software that link software to hardware. You don't have to use one. By the end of this course you will feel that for most of the apps we write we can use a general purpose OS like Linux or windows but sometime to be really efficient one might consider writing their own special purpose.

A lecture explaining the most critical hardware components and the OS,

CPU

Memory

Storage
Network

File System

Security

Program

Process

Kernel

Userspace

Kernel space (drivers)

System calls

In this lecture we show an example of the main function with instruction pointer moving to the next and execute. Simple with no stack or heap

The brilliant invention of the stack,  We introduce the stack function calling another function and local variables, return address, frames

It is understanding the stack that makes us pay attention to performance, too many function calls will jump from one to another

We Introduce the heap with an example, place where random blobs of memory is allocated, need to access it from multiple place and manipulate it. static vars are like this, except we can change the heap. Here lies the danger when it comes to threads as two threads changing the same place.

Allocating in the heap does not guarantee memory will be contagious, so if I store A and B , If I try to read A and B it might need two memory accesses as they might be in different pages, we will learn this in the virtual memory section. Whereas in the stack all variables are next to each other and memory access might give us both at once minimizing memory accesses.

Use linux, show all processes top

run ./a.out

then run cat /proc/29404/maps

this will list all mappings, including c library lib c, (except kernel mappings)

The process components

• We talk about memory, random access read and write memory, fast access volatile, DIMMs, double data read how CPU reads

We walk through the process of reading from memory, show how it must go through the CPU

We will discuss  the challenges of physical memory, we want to run many programs but we don't have enough memory contiguous should we just fail if I'm out of memory? Meet swap to disk. I have 10 programs all using libc a shared library, should I load 10 copies of the identical code or should I share them? I have 10 processes wanting to share memory.

We will talk cost of virtual memory, page tables  and is it really worth it

Internal fragmentation (memcached allocates large block but in the block lots of small space)

External fragmentation, the physical memory is available but all over the place cant even allocate a block to you

Shared memory

Isolation

Large Programs (Swap)

CPU involvement in memory reading is slow, Direct memory access allow a device to write or read directly between the device and user space as opposed to moving it from device to the register and write it back to memory

Going through Top command explain the memory list

Virtual memory

Residential

Shared

Code to allocate memory and leak, then fix it. and show memory

• Components

• ALU

• CU

• Registers and Caches

• Clock Speed

• Cores

• Pipelining

• Parallelism

• Architecture

• RISC vs CISC

• 32 bit 64 bit

uname -m to know the cpu architecture (32 or 64 bit)

cat /proc/cpuinfo

Walk through fetch an instruction, we check L1, L2, L3, if not, we get it  from memory, (64 byte cache line) put it in L3, L2 and L1i then we fetch the actual instruction to the IR (instruction register)

we then decode the instruction, covert the opcode of the instruction to actual instruction, RISC is simple, CISC takes time.  (CU)

Then we actually execute (ALU),

(optional memory read)

(write back to register)

we fetch another instruction, (cache hit L1)

Pipelining, one instruction requires many stages, fetch, decode, execute, memory access , memory write 

if we wait all stages cpu won't be doing anything, we pipeline them. The laundry example

Parallelism, multiple cores multiple processes simple

multiple cores single process, multiple threads

hyper threading or Simultaneous multi threading

- one core is represented as two logical core with logical registers (program counter etc) and caches are shared.

We discuss how we can saturate the workload of the process and thread

Mutexes and Semphores

Show a fork() and CoW where forking creates a new process

cow.c

cat /proc/1478/smaps | grep -A 50 heap

to grab the 50 lines after heap

notice the

Shared_Clean: 0 kB

Shared_Dirty: 9692 kB

Private_Clean: 0 kB

Private_Dirty: 552 kB

Why persist?

HDD vs SSD

App -> FileSystem -> Block Layer -> NVMe Driver

File system

FAT

ext4

Page cache

Read vs Write example

File Modes

Discuss some of the file modes O_SYNC, O_DIRECT

Little intro on NIC/ networks, OSI model, ip , tcp, UDP

socket

connection

Accept vs SYN queue

Receive vs Send Queue

tcp_hashinfo memory for mapping connection

select

epoll

IOCP

kqueue

io_uring

Build a tcp server with C 

Compiled Language vs Interrupted

Execution File format (exe/elf)

cgroup vs namespace