CS444/CS544 Operating Systems Threads 1/29/2007 Prof. Searleman jets@clarkson.edu

CS444/CS544 Spring 2007 Cooperating Processes Threads Reading assignment: Chapter 4 HW#2: system call

Recap: Processes A process includes How to create a process Address space (Code, Data, Heap, Stack) Register values (including the PC) Resources allocated to the process Memory, open files, network connections How to create a process Initializing the PCB and the address space (page tables) takes a significant amount of time Interprocess communication IPC is costly also Communication must go through OS (“OS has to guard any doors in the walls it builds around processes for their protection”)

Windows Process Creation BOOL CreateProcess( LPCTSTR lpApplicationName, // name of executable module LPTSTR lpCommandLine, // command line string LPSECURITY_ATTRIBUTES lpProcessAttributes, // SD LPSECURITY_ATTRIBUTES lpThreadAttributes, // SD BOOL bInheritHandles, // handle inheritance option DWORD dwCreationFlags, // creation flags LPVOID lpEnvironment, // new environment block LPCTSTR lpCurrentDirectory, // current directory name LPSTARTUPINFO lpStartupInfo, // startup information LPPROCESS_INFORMATION lpProcessInformation // process information );

Windows vs Unix Windows doesn’t maintain quite the same relationship between parent and child Later versions of Windows have concept of “job” to mirror UNIX notion of parent and children (process groups) Waiting for a process to complete? WaitforSingleObject to wait for completion GetExitCodeProcess ( will return STILL_ALIVE until process has terminated)

Cooperating Processes Processes can run independently of each other or processes can coordinate their activities with other processes To cooperate, processes must use OS facilities to communicate One example: parent process waits for child Many others Files (You’ve Used) Sockets (Networks) Pipes (Like Sockets for local machine; Pair of files) Signals (Today) Shared Memory Events Remote Procedure Call

Signals Processes can register to handle signals with the signal function void signal (int signum, void (*proc) (int)) Processes can send signals with the kill function kill (pid, signum) System defined signals like SIGHUP (0), SIGKILL (9), SIGSEGV(11) In UNIX shell, try: “kill –9 pidOfVictimProcess” Signals not used by system like SIGUSR1 and SIGUSR2 Note: sigsend/sigaction similar to kill/signal

Signals parentPid = getpid(); childPid = fork(); if (childPid == 0){ if (signal(SIGUSR1, sig_handler) == SIG_ERR) fprintf(stderr, "Unable to create handler for SIGUSR1\n"); if (signal(SIGUSR2, sig_handler) == SIG_ERR) fprintf(stderr, "Unable to create handler for SIGUSR2\n"); parentPid = getpid(); fprintf(stdout, "Parent process has id %d\n", parentPid); fprintf(stdout, "Parent process forks child...\n"); childPid = fork(); if (childPid == 0){ doChild(); } else { doParent(); }

doChild void doChild() { /* I am the child */ myPid = getpid(); assert(myPid != parentPid); fprintf(stdout, "In child (id %d) , Child process has id %d\n", myPid, myPid); /* send a SIG_USR1 to the parent */ fprintf(stdout, "Child process (id %d) sending 1st SIGUSR1 to parent process (id %d)\n", myPid, parentPid); err = kill(parentPid, SIGUSR1); if (err){ fprintf(stderr, "Child process (id %d) is unable to send SIGUSR1 signal to the parent process (id %d)\n", myPid, parentPid); }

doParent void doParent() { myPid = getpid(); assert(myPid == parentPid); fprintf(stdout, "In parent (id %d) , child process has id %d\n", myPid, childPid); fprintf(stdout, "Parent process (id %d) sending 1st SIGUSR2 to child process (id %d)\n", myPid, childPid); err = kill(childPid, SIGUSR2); if (err){ fprintf(stderr, "Parent process (id %d) is unable to send SIGUSR2 signal to the child process (id %d)\n", myPid, childPid); }

sigHandler static void sig_handler(int signo){ switch(signo){ case SIGUSR1: /* incoming SIGUSR1 signal */ handleSIGUSR1(); break; case SIGUSR2: /*incoming SIGUSR2 signal */ handleSIGUSR2(); case SIGTERM: /* incoming SIGTERM signal */ handleSIGTERM(); } return;

handleSIGUSR1 void handleSIGUSR1(){ numSIGUSR1handled++; if (myPid == parentPid){ fprintf(stdout, "Process %d: Parent Received SIGUSR1 %u\n", myPid, numSIGUSR1handled); } else { fprintf(stdout, "Error: Process %d: Received SIGUSR1, but I am not the parent!!\n", myPid); exit(1); } #if RECEIVE_MORE_THAN_ONE_SIGNAL if (signal(SIGUSR1, sig_handler) == SIG_ERR){ fprintf(stderr, "Unable to reset handler for SIGUSR1\n"); #endif

Sockets A socket is an end-point for communication over the network Create a socket int socket(int domain, int type, int protocol) Type = SOCK_STREAM for TCP Read and write socket just like files Can be used for communication between two processes on same machine or over the network

Pipes Bi-directional data channel between two processes on the same machine Created with: int pipe (int fildes[2]) Read and write like files

Remote Procedure Call (RPC) Marshalling objects

Problem which needs > 1 independent sequential process? Some problems are hard to solve as a single sequential process; easier to express the solution as a collection of cooperating processes Hard to write code to manage many different tasks all at once How would you write code for “make phone calls while making dinner while doing dishes while looking through the mail” Can’t be independent processes because share data (your brain) and share resources (the kitchen and the phone) Can’t do them sequentially because need to make progress on all tasks at once Easier to write “algorithm” for each and when there is a lull in one activity let the OS switch between them On a multiprocessor, exploit parallelism in problem

Example: Web Server Web servers listen on an incoming socket for requests Once it receives a request, it ignores listening to the incoming socket while it services the request Must do both at once One solution: Create a child process to handle the request and allow the parent to return to listening for incoming requests Problem: This is inefficient because of the address space creation (and memory usage) and PCB initialization

Observation There are similarities in the process that are spawned off to handle requests They share the same code, have the same privileges, share the same resources (html files to return, cgi script to run, database to search, etc.) But there are differences Operating on different requests Each one will be in a different stage of the “handle request” algorithm

Idea Let these tasks share the address space, privileges and resources Give each their own registers (like the PC), their own stack etc Process – unit of resource allocation (address space, privileges, resources) Thread – unit of execution (PC, stack, local variables)

Single-Threaded vs Multithreaded Processes

Process vs Thread Each thread belongs to one process One process may contain multiple threads Threads are logical unit of scheduling Processes are the logical unit of resource allocation

Address Space Map For Single-Threaded Process Biggest Virtual Address Stack (Space for local variables etc. For each nested procedure call) Stack Pointer Heap (Space for memory dynamically allocated e.g. with malloc) Statically declared variables (Global variables) Code (Text Segment) PC Ox0000

Address Space Map For Multithreaded Process Biggest Virtual Address Thread 1 stack SP (thread 1) Thread 2 stack SP (thread 2) Heap (Space for memory dynamically allocated e.g. with malloc) Statically declared variables (Global variables) Code (Text Segment) PC (thread 2) PC (thread 1) Ox0000

Kernel support for threads? Some OSes support the notion of multiple threads per process and others do not Even if no “kernel threads” can build threads at user level Each “multi-threaded” program gets a single kernel in the process During its timeslice, it runs code from its various threads User-level thread package schedules threads on the kernel level process much like OS schedules processes on the CPU SAT question? CPU is to OS is to processes like? Kernel thread is to User-level thread package is to user threads User-level thread switch must be programmed in assembly (restore of values to registers, etc.)

User-level Threads

User-level threads How do user level thread packages avoid having one thread monopolize the processes time slice? Solve much like OS does Solution 1: Non-preemptive Rely on each thread to periodically yield Yield would call the scheduling function of the library Solution 2: OS is to user level thread package like hardware is to OS Ask OS to deliver a periodic timer signal Use that to gain control and switch the running thread

Kernel vs User Threads One might think that kernel level threads are best, and only if the kernel does not support threads should you use user level threads In fact, user level threads can be much faster Thread creation , “Context switch” between threads, communication between threads all done at user level Procedure calls instead of system calls (verification of all user arguments, etc.) in all these cases!

Problems with User-level threads OS does not have information about thread activity and can make bad scheduling decisions Examples: If thread blocks, whole process blocks Kernel threads can overlap I/O and computation within a process! Kernel may schedule a process with all idle threads

Scheduler Activations If kernel level thread support is available, then use both kernel threads *and* user-level threads Each process requests a number of kernel threads to use for running user-level threads on Kernel promises to tell user-level before it blocks a kernel thread so user-level thread package can choose what to do with the remaining kernel level threads User level promises to tell kernel when it no longer needs a given kernel level thread

Thread Support Pthreads is a user-level thread library Can use multiple kernel threads to implement it on platforms that have kernel threads Java threads (extend Thread class) run by the Java Virtual Machine Kernel threads Linux has kernel threads (each has its own task_struct), created with clone system call Each user level thread maps to a single kernel thread (Windows 95/98/NT/2000/XP, OS/2) Many user level threads can map onto many kernel level threads like scheduler activations (Windows NT/2000 with ThreadFiber package, Solaris 2)

Pthreads Interface POSIX threads, user-level library supported on most UNIX platforms Much like the similarly named process functions thread = pthread_create(procedure) pthread_exit pthread_wait(thread) Note: To use pthreads library, #include <pthread.h> compile with -lpthread

Pthreads Interface (cont.) Pthreads support a variety of functions for thread synchronization/coordination Used for coordination of threads (ITC) – more on this soon! Examples: Condition Variable(pthread_cond_wait, pthread_signal) Mutex(pthread_mutex_lock, pthread_mutex_unlock)

Performance Comparison Processes Fork/Exit 251 Kernel Threads Pthread_create/ Pthread_join 94 User-level Threads 4.5 In microseconds, on a 700 MHz Pentium, Linux 2.2.16, Steve Gribble, 2001.

Windows Threads HANDLE CreateThread( LPSECURITY_ATTRIBUTES lpThreadAttributes, DWORD dwStackSize, LPTHREAD_START_ROUTINE lpStartAddress, DWORD dwCreationFlags, LPVOID lpParameter, LPDWORD lpThreadId);

Windows Thread Synchronization Windows supports a variety of objects that can be used for thread synchronization Examples Events (CreateEvent, SetEvent, ResetEvent, WaitForSingleObject) Semaphores (CreateSemaphore, ReleaseSemaphore, WaitForSingleObject) Mutexes (CreateMutex, ReleaseMutex, WaitForSingleObject) WaitForMultipleObject More on this when we talk about synchronization

Warning: Threads may be hazardous to your health One can argue (and John Ousterhout did) that threads are a bad idea for most purposes Anything you can do with threads you can do with an event loop Remember “make phone calls while making dinner while doing dishes while looking through the mail” Ousterhout says thread programming to hard to get right

Outtakes Processes that just share code but do not communicate Wasteful to duplicate Other ways around this than threads

Example: User Interface Allow one thread to respond to user input while another thread handles a long operation Assign one thread to print your document, while allowing you to continue editing

Benefits of Concurrency Hide latency of blocking I/O without additional complexity Without concurrency Block whole process Manage complexity of asynchronous I/O (periodically checking to see if it is done so can finish processing) Ability to use multiple processors to accomplish the task Servers often use concurrency to work on multiple requests in parallel User Interfaces often designed to allow interface to be responsive to user input while servicing long operations

Thread pools What they are and how they avoid thread creation overhead

Experiment Start up various processes under Windows (Word, IE,..) How many processes are started? How many threads and of what priority?

Scheduling CPU or “short term” scheduler selects process from ready queue (every 10 msec or so) “dispatcher” does the process switching “long-term” scheduler controls “degree of multiprogramming” (number of processes in memory); selects a good “job mix” “job mix” – I/O-bound, CPU-bound, interactive, batch, high priority, background vs. foreground, real-time “non-preemptive” (cooperative) vs. “preemptive”

Performance Measures Throughput: #processes/time unit Turnaround time: time completed – time submitted Waiting time: sum of times spent in ready queue Response time: time from submission of a request until the first response is produced Variation of response time (predictability) CPU utilization Disk (or other I/O device) utilization

I/O-bound & CPU-bound Device1 P1 CPU time quantum Device2 P2 CPU

I/O-bound & CPU-bound P1: CPU-bound Device1 idle Device1 idle CPU idle CPU idle Turnaround time for P1

I/O-bound & CPU-bound P2: I/O-bound Device2 idle Device2 idle CPU idle Turnaround time for P2

I/O-bound & CPU-bound Schedule1: non-preemptive, P1 selected first Turnaround time for P1 Turnaround time for P2 Without P1

I/O-bound & CPU-bound Schedule2: non-preemptive, P2 selected first Turnaround time for P1 Turnaround time for P2

I/O-bound & CPU-bound How does the OS know whether a process is I/O-bound or CPU-bound? can monitor the behavior of a process & save the info in the PCB example: how much CPU time did the process use in its recent time quanta? (a small fraction => I/O intensive; all of the quantum => CPU intensive) The nature of a typical process changes from I/O-bound to CPU-bound and back as it works through its Input/Process/Output Cycle

Preemptive vs. Non-Preemptive ready: P1, P2 t1 ready: P2 blocked: P1 t2

Preemptive vs. Non-Preemptive Non-Preemptive: must continue to run P1 at t3 Preemptive: can choose between P1 & P2 at t3 t2 ready: P1 blocked: P2 t3 ready: P2 running: P1

nonpreemptive (cooperative): (1) and (4) only preemptive: otherwise New Ready Running Waiting Terminated admit dispatch (4) exit, abort (2) interrupt (1) block for I/O or wait for event (3) I/O completed or event occurs nonpreemptive (cooperative): (1) and (4) only preemptive: otherwise