mirror of
https://github.com/jackyzha0/quartz.git
synced 2025-12-28 07:14:05 -06:00
105 lines
3.7 KiB
Markdown
105 lines
3.7 KiB
Markdown
---
|
||
title: "22-virtual-memory"
|
||
aliases:
|
||
tags:
|
||
- cosc204
|
||
- lecture
|
||
---
|
||
|
||
# Swapping and virtual memory
|
||
swapping
|
||
- if there is not enough physical memory we need to sawp processes out of the main ememory to the secondary storage e.g., disk
|
||
- 
|
||
- when a process is ready, it is swapped into the main memory
|
||
- allows more processes to multitask
|
||
|
||
partially loaded proceses
|
||
- dynamic loading
|
||
- load a potion of code when it is called as some code may not need to be executed, e.g., code for handling errors
|
||
- pros
|
||
- process not limited by amount of avilable memory
|
||
- more processes multitasking
|
||
- quicker to swap than entire process
|
||
|
||
virtual memory
|
||
- idea that processes dont need to be fully in memory to run
|
||
|
||
extends main memory to secondary storage, and allows dynamic loading of processes while they execute
|
||
- programmer deals with vmem just like paging scheme
|
||
- mem manager in OS kernel controls loading pages of the process into main mem from secondary storage
|
||
|
||
# Demand paging
|
||
dont load a page into mem until it is referenced by CPU
|
||
|
||
implementing
|
||
- in paging scheme there are extra bits in the table to provide more information
|
||
- valid/invlid bit
|
||
- protection bits
|
||
- in demand paging the mem manager uses valid/invalid bit to tell if a page is loaded
|
||
|
||
|
||
|
||
note: each page can be stored twice: loaded in main memory and in the backing store - these two copies need to remain consistent. i.e.., changes to one must be reflected in the other
|
||
|
||
page faults
|
||
- "trap" occurs when trying to access and invalid page
|
||
- what the os does:
|
||
- check if caused by invalid memory access or unavailable page frame
|
||
- if nvalid mem access
|
||
- terminate process
|
||
- else
|
||
- find a free frame
|
||
- read the page from the disk to the free frame
|
||
- modify the page table
|
||
- restart the instruction
|
||
- VMAs are searched to find valid areas or memory access
|
||
|
||
# page replacement algorithms
|
||
If main memory is full when a page fault is generated, one of the pages currently being held needs to be replaced.
|
||
|
||
This means an extra step in the operating system’s page-servicing routine.
|
||
- Find the desired page on the backing store (secondary storage).
|
||
- Find a free frame of memory
|
||
- If there’s a free frame in memory, use it
|
||
- Otherwise: Select a frame to swap out
|
||
- Save it to the backing store (in case it’s changed)
|
||
- Proceed as before.
|
||
|
||
FIFO
|
||
- replace the oldest page
|
||
- 
|
||
- pro
|
||
- simple to understand/implement
|
||
- con
|
||
- maybe first page to be ref'd is often being referenced
|
||
- belady's anomaly: possible to increase numbe of page faults when increasing the number of frames in memory
|
||
|
||
Optimal page replacement
|
||
- replace the page that will not be used for the longest time
|
||
- 
|
||
- pro: optimal
|
||
- con: can predict future
|
||
|
||
LRU (least recently used)
|
||
- replace page than has not been used for the longest time
|
||
- 
|
||
- pro: optimal if you look back in time
|
||
- con: time consuming to keep record of LRU
|
||
- future memory references resemble past ones
|
||
|
||
# frame allocation
|
||
how many frames to allocate to a process?
|
||
|
||
how to distribute frames among n processes in a multitasking scenario
|
||
- equal allocation: all created equal: so each get equal number of frames
|
||
- proportional allocation: allocate depending on size of process
|
||
- local vs global replacement:
|
||
- local: steal from your own frames
|
||
- global: can steal from "victim" processes
|
||
|
||
# thrashing
|
||
when a proces has too lttle frames it will have a lot of page faults.
|
||
|
||
a thrashing process spending more time paging than executing
|
||
|
||
the working set of a process is the number of pages a process needs in orde |