EAGER: Efficient Monitor-Based Synchronization Mechanisms for Concurrent Programs

点击翻译按钮获取中文摘要

Programming multicore processors is a challenging task due to bugs resultingfrom concurrency and synchronization. A fundamental reason for the difficultyis that current programming mechanisms require programmers to manuallycoordinate synchronization between threads running on these systems.Consequently, concurrent systems are prone to bugs due to faultysynchronization problems such as missing, wrong, or lost notifications forthreads. Moreover, these bugs are hard to find during testing because they mayappear only under a rare schedule of events. Current programming mechanismsalso limit the amount of concurrency in threads to ensure that a sharedvariable is not updated by multiple threads. As a result, many concurrentsystems slow down due to sequential bottleneck of accessing shared variables.This project is expected to fundamentally change the way synchronizationmechanisms (such as monitors) are written and implemented. It will lead tobetter understanding of how conditions in a multithreaded program can beevaluated efficiently. As a result, the concurrent programming systems willbecome more reliable and faster. The impact on society is expected to be largebecause all new systems are based on concurrent programs.The project is developing new synchronization mechanisms with two fundamentalgoals. The first goal is to make synchronization as simple and intuitive aspossible for programmers. Current (monitor based) synchronizationmechanisms require programmers to explicitly signal threads that may be waitingon certain conditions. In the synchronization mechanism developed in thisproject, there is no notion of condition variables and it is the responsibilityof the runtime system to automatically signal appropriate threads. Analogousto automatic garbage collection in Java, the project is designing efficientalgorithms and techniques for automatic signaling. The project isinvestigating techniques to deal with fairness and exploitation of runtimeinformation such as idle cores for effective and efficient automatic signaling.The second goal is to increase the concurrency in implementation of monitorwithout requiring programmers to do any additional work and without violatingthe guarantee provided by the monitor that execution of its methods isequivalent to one in which there is at most one thread in the monitor at anytime. There is also a strong educational component for this project. Synchronization is a fundamental topicin computer science education and is taught in all operating system courses. Byeliminating manual signaling from these courses, students would get more timeto focus on higher level concurrent programming concerns.