Chapter 2: Processes & Threads. Chapter 2

Transcription

1 : Processes & Threads

2 Processes ad threads Processes Threads Schedulig Iterprocess commuicatio Classical IPC problems (origialy modified by Etha 2

3 What is a process? Code, data, ad stack Usually (but ot always) has its ow address space Program state CPU registers Program couter (curret locatio i the code) Stack poiter Oly oe process ca be ruig i the CPU at ay give time! (origialy modified by Etha 3

4 The process model Sigle PC (CPU s poit of view) A B C B D A Multiple PCs (process poit of view) B C D Multiprogrammig of four programs Coceptual model 4 idepedet processes Processes ru sequetially Oly oe program active at ay istat! That istat ca be very short D C B A Time (origialy modified by Etha 4

5 Whe is a process created? Processes ca be created i two ways System iitializatio: oe or more processes created whe the OS starts up Executio of a process creatio system call: somethig explicitly asks for a ew process System calls ca come from User request to create a ew process (system call executed from user shell) Already ruig processes User programs System daemos (origialy modified by Etha 5

6 Process Creatio 6 Paret process creates childre processes, which, i tur create other processes, formig a tree of processes Geerally, process idetified ad maaged via a process idetifier (pid) Resource sharig optios Paret ad childre share all resources Childre share subset of paret s resources Paret ad child share o resources Executio optios Paret ad childre execute cocurretly Paret waits util childre termiate Sprig 2018 CS/ CO E Ope rati g Syst ems She rif Kha ttab

7 Process Creatio (Cot.) Address space Child duplicate of paret Child has a program loaded ito it UNIX examples fork() system call creates ew process exec() system call used after a fork() to replace the process memory space with a ew program 7 Sprig 2018 CS/ CO E Ope rati g Syst ems She rif Kha ttab

8 Whe do processes ed? Coditios that termiate processes ca be Volutary Ivolutary Volutary Normal exit Error exit Ivolutary Fatal error (oly sort of ivolutary) Killed by aother process (origialy modified by Etha 8

9 Process Termiatio 9 Process executes last statemet ad the asks the operatig system to delete it usig the exit() system call. Returs status data from child to paret (via wait()) Process resources are deallocated by operatig system Paret may termiate the executio of childre processes usig the abort() system call. Some reasos for doig so: Child has exceeded allocated resources Task assiged to child is o loger required Syst The paret is exitig ad the operatig systems does ot ems allow a child to cotiue if its paret termiates Sprig 2018 CS/ CO E Ope rati g She rif Kha ttab

10 Process Termiatio 10 Some operatig systems do ot allow a child to exist if its paret has termiated. If a process termiates, the all its childre must also be termiated. cascadig termiatio. All childre, gradchildre, etc. are termiated. The termiatio is iitiated by the operatig system. The paret process may wait for termiatio of a child process by usig the wait()system call. The call returs status iformatio ad the pid of the termiated process pid = wait(&status); E If o paret waitig (did ot ivoke wait()) process is a zombie Ope rati g If paret termiated without ivokig wait, process is Syst a orpha Sprig 2018 CS/ CO ems She rif Kha ttab

11 Process hierarchies Paret creates a child process Child processes ca create their ow childre Forms a hierarchy UNIX calls this a process group If a process exits, its childre are iherited by the exitig process s paret Widows has o cocept of process hierarchy All processes are created equal (origialy modified by Etha 11

12 Process states 5 Blocked (waitig) 7 Created 1 Ready Exit 2 Ruig 6 Process i oe of 5 states Created Ready Ruig Blocked Exit Trasitios betwee states 1 - Process eters ready queue 2 - Scheduler picks this process 3 - Scheduler picks a differet process 4 - Process waits for evet (such as I/O) 5 - Evet occurs 6 - Process exits 7 - Process eded by aother process (origialy modified by Etha 12

13 Processes i the OS Two layers for processes Lowest layer of process-structured OS hadles iterrupts, schedulig Above that layer are sequetial processes Processes tracked i the process table Each process has a process table etry Processes 0 1 N-2 N-1 Scheduler (origialy modified by Etha 13

14 What s i a process table etry? May be stored o stack Process maagemet Registers Program couter CPU status word Stack poiter Process state Priority / schedulig parameters Process ID Paret process ID Sigals Process start time Total CPU usage File maagemet Root directory Workig (curret) directory File descriptors User ID Group ID Memory maagemet Poiters to text, data, stack or Poiter to page table (origialy modified by Etha 14

15 What happes o a trap/iterrupt? 1. Hardware saves program couter (o stack or i a special register) 2. Hardware loads ew PC, idetifies iterrupt 3. Assembly laguage routie saves registers 4. Assembly laguage routie sets up stack 5. Assembly laguage calls C to ru service routie 6. Service routie calls scheduler 7. Scheduler selects a process to ru ext (might be the oe iterrupted ) 8. Assembly laguage routie loads PC & registers for the selected process (origialy modified by Etha 15

16 Threads: processes sharig memory Process == address space Thread == program couter / stream of istructios Two examples Three processes, each with oe thread Oe process with three threads Process 1 Process 2 Process 3 Process 1 User space System space Threads Kerel Threads Kerel (origialy modified by Etha 16

17 Process & thread iformatio Per process items Address space Ope files Child processes Sigals & hadlers Accoutig ifo Global variables Per thread items Program couter Registers Stack & stack poiter State Per thread items Program couter Registers Stack & stack poiter State Per thread items Program couter Registers Stack & stack poiter State (origialy modified by Etha 17

18 Threads & Stacks Thread 1 Thread 2 Thread 3 User space Thread 1 s stack Process Thread 3 s stack Thread 2 s stack Kerel => Each thread has its ow stack! (origialy modified by Etha 18

19 Why use threads? Allow a sigle applicatio to do may thigs at oce Simpler programmig model Less waitig Threads are faster to create or destroy No separate address space Overlap computatio ad I/O Could be doe without threads, but it s harder Example: word processor Thread to read from keyboard Thread to format documet Thread to write to disk Whe i the Course of huma evets, it becomes ecessary for oe people to dissolve the political bads which have coected them with aother, ad to assume amog the powers of the earth, the separate ad equal statio to which the Laws of Nature ad of Nature's God etitle them, a decet respect to the opiios of makid requires that they should declare the causes which impel them to the separatio. We hold these truths to be self-evidet, that all me are created equal, that they are edowed by their Creator with certai ualieable Rights, that amog these are Life, Liberty ad the pursuit of Happiess.--That to secure these rights, Govermets are istituted amog Me, derivig their just powers from the coset of the govered, --That wheever ay Form of Govermet becomes Kerel destructive of these eds, it is the Right of the People to alter or to abolish it, ad to istitute ew Govermet, layig its foudatio o such priciples ad orgaizig its powers i such form, as to them shall seem most likely to effect their Safety ad Happiess. Prudece, ideed, will dictate that Govermets log established should ot be chaged for light ad trasiet causes; ad accordigly all (origialy modified by Etha 19

20 Multithreaded Web server Dispatcher thread Worker thread while(true) { getnextrequest(&buf); hadoffwork(&buf); } Network coectio Kerel Web page cache while(true) { waitforwork(&buf); lookforpageicache(&buf,&page); if(pagenoticache(&page)) { readpagefromdisk(&buf,&page); } returpage(&page); } (origialy modified by Etha 20

21 Three ways to build a server Thread model Parallelism Blockig system calls Sigle-threaded process: slow, but easier to do No parallelism Blockig system calls Fiite-state machie Each activity has its ow state States chage whe system calls complete or iterrupts occur Parallelism Noblockig system calls Iterrupts (origialy modified by Etha 21

22 Implemetig threads Process Thread Kerel Kerel Ru-time system Thread table Process table Process table Thread table User-level threads + No eed for kerel support - May be slower tha kerel threads - Harder to do o-blockig I/O Kerel-level threads + More flexible schedulig + No-blockig I/O - Not portable (origialy modified by Etha 22

23 Iterprocess Commuicatio 23 Processes withi a system may be idepedet or cooperatig Cooperatig process ca affect or be affected by other processes, icludig sharig data Reasos for cooperatig processes: Iformatio sharig Computatio speedup Modularity Coveiece Cooperatig processes eed iterprocess commuicatio (IPC) Two models of IPC Shared memory Message passig Sprig 2018 CS/ CO E Ope rati g Syst ems She rif Kha ttab

24 Producer-Cosumer Problem Paradigm for cooperatig processes, producer process produces iformatio that is cosumed by a cosumer process 24 Sprig 2018 CS/ CO E Ope rati g Syst ems She rif Kha ttab

25 Bouded-Buffer Shared-Memory Solutio Shared data #defie BUFFER_SIZE 10 typedef struct {... } item; 25 item buffer[buffer_size]; it i = 0; it out = 0; Syst Solutio is correct, but ca oly use BUFFER_SIZE- ems 1 elemets Sprig 2018 CS/ CO E Ope rati g She rif Kha ttab

26 Bouded-Buffer Producer item ext_produced; while (true) { /* produce a item i ext produced */ while (((i + 1) % BUFFER_SIZE) == out) ; /* do othig */ buffer[i] = ext_produced; i = (i + 1) % BUFFER_SIZE; } 26 Sprig 2018 CS/ CO E Ope rati g Syst ems She rif Kha ttab

27 Bouded Buffer Cosumer item ext_cosumed; while (true) { while (i == out) ; /* do othig */ ext_cosumed = buffer[out]; out = (out + 1) % BUFFER_SIZE; /* cosume the item i ext ems cosumed */ 27 Sprig 2018 CS/ CO E Ope rati g Syst She rif Kha ttab

28 Schedulig What is schedulig? Goals Mechaisms Schedulig o batch systems Schedulig o iteractive systems Other kids of schedulig Real-time schedulig (origialy modified by Etha 28

29 Why schedule processes? Bursts of CPU usage alterate with periods of I/O wait Some processes are CPU-boud: they do t make may I/O requests Other processes are I/O-boud ad make may kerel requests CPU boud Total CPU usage CPU bursts I/O waits I/O boud Time (origialy modified by Etha Total CPU usage 29

30 Whe are processes scheduled? At the time they eter the system Commo i batch systems Two types of batch schedulig Submissio of a ew job causes the scheduler to ru Schedulig oly doe whe a job volutarily gives up the CPU (i.e., while waitig for a I/O request) At relatively fixed itervals (clock iterrupts) Necessary for iteractive systems May also be used for batch systems Schedulig algorithms at each iterrupt, ad picks the ext process from the pool of ready processes (origialy modified by Etha 30

31 Schedulig goals All systems Fairess: give each process a fair share of the CPU Eforcemet: esure that the stated policy is carried out Balace: keep all parts of the system busy Batch systems Throughput: maximize jobs per uit time (hour) Turaroud time: miimize time users wait for jobs CPU utilizatio: keep the CPU as busy as possible Iteractive systems Respose time: respod quickly to users requests Proportioality: meet users expectatios Real-time systems Meet deadlies: missig deadlies is a system failure! Predictability: same type of behavior for each time slice (origialy modified by Etha 31

32 Measurig schedulig performace Throughput Amout of work completed per secod (miute, hour) Higher throughput usually meas better utilized system Respose time Respose time is time from whe a commad is submitted util results are retured Ca measure average, variace, miimum, maximum, May be more useful to measure time spet waitig Turaroud time Like respose time, but for batch jobs (respose is the completio of the process) Usually ot possible to optimize for all metrics with the same schedulig algorithm (origialy modified by Etha 32

33 First Come, First Served (FCFS) Curret job queue A B C D FCFS scheduler Executio order A B C D Goal: do jobs i the order they arrive Fair i the same way a bak teller lie is fair Simple algorithm! Problem: log jobs delay every job after them May processes may wait for a sigle log job (origialy modified by Etha 33

34 Shortest Job First (SJF) A B C D SJF scheduler B Curret job queue Executio order D A C Goal: do the shortest job first Short jobs complete first Log jobs delay every job after them Jobs sorted i icreasig order of executio time Orderig of ties does t matter Shortest Remaiig Time First (SRTF): preemptive form of SJF Problem: how does the scheduler kow how log a job will take? (origialy modified by Etha 34

35 Three-level schedulig CPU CPU scheduler Arrivig jobs Iput queue Admissio scheduler Mai memory Memory scheduler Jobs held i iput queue util moved ito memory Pick complemetary jobs : small & large, CPU- & I/O-itesive Jobs move ito memory whe admitted CPU scheduler picks ext job to ru Memory scheduler picks some jobs from mai memory ad moves them to disk if isufficiet memory space (origialy modified by Etha 35

36 Roud Robi (RR) schedulig Roud Robi schedulig Give each process a fixed time slot (quatum) Rotate through ready processes Each process makes some progress What s a good quatum? Too short: may process switches hurt efficiecy Too log: poor respose to iteractive requests Typical legth: ms E D C B A A B C D E Time (origialy modified by Etha 36

37 Priority schedulig Assig a priority to each process Ready process with highest priority allowed to ru Ruig process may be iterrupted after its quatum expires Priorities may be assiged dyamically Reduced whe a process uses CPU time Icreased whe a process waits for I/O Ofte, processes grouped ito multiple queues based o priority, ad ru roud-robi per queue High Ready processes Priority 4 Priority 3 Priority 2 Priority 1 Low (origialy modified by Etha 37

38 Shortest process ext Ru the process that will fiish the sooest I iteractive systems, job completio time is ukow! Guess at completio time based o previous rus Update estimate each time the job is ru Estimate is a combiatio of previous estimate ad most recet ru time Not ofte used because roud robi with priority works so well! (origialy modified by Etha 38

39 Lottery schedulig Give processes tickets for CPU time More tickets => higher share of CPU Each quatum, pick a ticket at radom If there are tickets, pick a umber from 1 to Process holdig the ticket gets to ru for a quatum Over the log ru, each process gets the CPU m/ of the time if the process has m of the existig tickets Tickets ca be trasferred Cooperatig processes ca exchage tickets Cliets ca trasfer tickets to server so it ca have a higher priority (origialy modified by Etha 39

40 Policy versus mechaism Separate what may be doe from how it is doe Mechaism allows Priorities to be assiged to processes CPU to select processes with high priorities Policy set by what priorities are assiged to processes Schedulig algorithm parameterized Mechaism i the kerel Priorities assiged i the kerel or by users Parameters may be set by user processes Do t allow a user process to take over the system! Allow a user process to volutarily lower its ow priority Allow a user process to assig priority to its threads (origialy modified by Etha 40

41 Schedulig user-level threads Process A Ru-time system Thread table Kerel Process B Process table Kerel picks a process to ru ext Ru-time system (at user level) schedules threads Ru each thread for less tha process quatum Example: processes get 40ms each, threads get 10ms each Example schedule: A1,A2,A3,A1,B1,B3,B2,B3 Not possible: A1,A2,B1,B2,A3,B3,A2,B1 (origialy modified by Etha 41

42 Schedulig kerel-level threads Process A Kerel Process table Process B Thread table Kerel schedules each thread No restrictios o orderig May be more difficult for each process to specify priorities Example schedule: A1,A2,A3,A1,B1,B3,B2,B3 Also possible: A1,A2,B1,B2,A3,B3,A2,B1 (origialy modified by Etha 42