Your task in this project is to write a non-preemptive user-level threads package. Most operating systems, e.g. Solaris, Mach and DEC OSF, provide such a lightweight thread abstraction because, as you might recall from the lectures and the reading, user-level threads provide concurrency with very low-overhead. NT does not have a user-level threads library. Your task is to write one.
We have provided you with some skeletal code that creates an environment on top of NT that closely resembles the environment on top of raw hardware shortly after booting. We use NT to provide us with a virtual processor and memory. NT also bootstraps our kernel; that is, it loads it into memory and starts executing it at its entry point. Your task is to build high-level abstractions on top of this environment in the same manner NT builds its abstractions on top of the hardware.
There are a few distinct components to this project.
First, you will have to write some generic FIFO (first-in, first-out) enqueue and dequeue operations. We will be relying on this queue implementation throughout the rest of the semester, so it's important that the implementation be efficient. Specifically, enqueue and dequeue operations should both work in O(1) time.
Second, you need to define and implement thread control blocks, and the functions that operate on them. We suggest you start with minithread_create and minithread_yield, and go on to implement the scheduler. Once you have those two working, you can come back to implement the rest of the functionality.
Third, you need to implement a scheduler. For this assignment, all you need is a first-come, first-served scheduler. You can assume that your threads will voluntarily give up the CPU, which means that your test programs should make occasional calls to minithread_yield().
Fourth, you need to implement semaphores in order to be able to synchronize multiple threads.
Finally, you need to demonstrate your threads package by implementing a solution to the "food services" problem. This is actually an instance of the better-known multiple-producer, multiple-consumer problem. There are N cooks (each a separate minithread) that constantly produce burgers. We'll assume for debugging purposes that each burger has a unique id assigned to it at the time it is created. The cooks place their burgers on a stack (at the front of the queue you built earlier in this assignment). There are M hungry students (each a separate minithread) that constantly grab burgers from the stack and eat them. Each time a student eats a burger, she prints a message containing the id of the burger she just ate. Ensure, for health and sanity reasons, that a given burger is consumed at most once by at most one student.
If you decide not to use the integrated development environment (visual studio), you can use any other editor you like and compile the code from the command line. To compile the code from the command line, go to the directory where you unpacked the project and type make to compile the project incrementally (will detect which files have been modified and recompile them) or nmake clean to delete all of the intermediate files. Advanced students wishing to use the Jornadas early can do a nmake wince once they have all of the necessary software installed, and run their code on the Jornadas. Note that for this assignment, your code does not need to run on the Jornadas - we'll start to use these devices starting with the next project.
Three small test programs, test1.c, test2.c, and test3.c, are included in the minithreads.zip file. They test spawning a thread, spawning multiple threads, and ping-ponging between threads. They should be useful when you are implementing minithreads. Two larger test programs, buffer.c, a bounded buffer, and sieve.c, a sieve for finding primes, are more complicated. Ideally, you should test your minithreads implementation against these.
If you have problems with the project or questions, please send mail to cs414@cs.cornell.edu or come to office hours for one of the TAs.
It's crucial that systems code be correct and robust. You must test your code with reasonable and unreasonable test cases, and ensure that it behaves correctly. Note that you should maintain some separation between the minithread package and minithread applications. Most notably, your minithread applications should not contain any dependencies on the scheduling algorithm or on the lack of preemption.
To facilitate testing, we provided you with some test programs. It's a good idea to start with these, and develop your own tests as you need them. The simplest test cases are test1.c, test2.c and test3.c, which test single thread creation, multiple thread creation, and ping-pong between two threads. buffer.c provides a bounded buffer implementation. A producer and consumer keep producing values and consuming them across a buffer of finite length. sieve.c is a Sieve of Eratosthenes for concurrently searching for primes. It has a single thread on one end, injecting the numbers 1, 2, 3, 4, 5, 6, 7, 8, ... into a pipe. For each prime p, there is a thread in the middle of the pipe that consumes a number from the pipe if that number is divisible by p. Otherwise, it passes the value on to the next thread in the pipe. At the very end, there is a thread that prints the values that emerge from the pipe. Note that this assembly will only print out prime numbers, because the threads in the pipe will consume all non-primes.
Since we will soon make the threads package preemptive, all code you write should be properly synchronized. That is, everything should work no matter where in the application thread you place a minithread_yield. Consequently, it's a good idea to test your code with minithread_yield's inserted at random locations throughout the application code (note that we don't expect the system code in minithread.c or synch.c to be yield-safe for this project - just the applications).
Do not forget to check for memory leaks. Your threads package should not run out of memory when large numbers of threads are created and destroyed.
Here are the frequently asked questions on project 1.
For this project you have to submit the source of all the files you modify (at least minithread.c, queue.c, synch.c, consprod.c) and the NT executable for the multi-producer, multi-consumer problem. Also submit the Makefile with the MAIN symbol set to the name of your producer-consumer program.
Implement the barber shop problem. That is:
Add preemption and implement a multilevel feedback queue with four
levels, with round-robin at each level, and where the time quanta
doubles at every level. Demonstrate that it works as intended.
Final Word
If you need help with any part of the assignment, we are
here to help. If you start
early, this should be an incredibly fun project.