Download Lecture Notes on Threads and Processes - Operating Systems | CMPSCI 377 and more Study notes Operating Systems in PDF only on Docsity!
CMPSCI 377 Operating Systems Fall 2005
Lecture 5: September 27
Lecturer: Emery Berger Scribe: Gabor Asztalos, Kat Reagan
Today:
- Intro to threads and their distinction/relation to a process
5.1 Threads
5.1.1 Threads and Processes
Threads are a subclass of processes. Process(Recall): -Is a unit of execution. -Control, address space, resources. -Call fork() to create. Thread: -Control only aka pc, stack, registers. -Created with pthreadcreate(). -A process can contain many threads becuase threads are finer grain(see below). -Threads communicate through memory yielding fastest possible communication.
Definition: Granularity - description of ’size’. Fine grain = small, coarse grain = large.
5.1.2 Thread System History/Evolution
MS-Dos: -No threads. -One process. -Result:Nothing could be done simultaneously. Forced to wait for termination of process even for simple I/O. Embedded Systems: -One process. -Many threads. -Result:Good structure for these systems for there are usually many events to recognize but very simple responses. Unix, Ultrix, MacOS(preX): -Many processes.
5-2 Lecture 5: September 27
-At most one thread per process. -Result:Finally your 8 mhz cpu could run more than one process, but these processes could not do much at once. Today: -Many processes. -Each with many threads. -Result:Your disk can’t keep up with the processor.
5.1.3 Types of Threads
Kernel Threads: -Light weight processes. -Scheduled by OS. -Context switch required. -Involves the OS meaning time sharing(quantas). -Result: -Can be scheduled on multiple processors. -Threads can mask latency Because don’t block on I/O. aka waiting to read from device. -Can be slow, but still faster than processes.
User-Level Threads: -Zero OS invlolvement. -Process containing thread is all that is known. ex: java green threads. Threads have own scheduler. -No context switch. -Result: -Flexible, easy scheduling. -No system calls = fast. -Requires cooperative threads. One uncoooperative thread can take over. -OS only sees processes not threads. Threads block on I/O. Restricted to one processor.
Hybird Method: -Uses mapping to separate and abstract. -Kernel Thread ===- LWP’s ===- user lever threads. -Result: -If there is a fault in user level thread like a block on I/O it creates another LWP to handle the thread. This, however, requires load balancing(see below). -Best of both worlds, but never made it due to confussions.
Load Balancing - there are numerous processes on many processors each with many threads on todays machines. How can we best distribute the workload in the form of these threads between the processes?
