عنوان پایان‌نامه
مدیریت دما از طریق تخصیص و زمانبندی وظایف در سیستم های چند هسته ای بر روی یک تراشه
- رشته تحصیلی
- مهندسی کامپیوتر-معماری کامپیوتر
- مقطع تحصیلی
- کارشناسی ارشد
- محل دفاع
- کتابخانه مرکزی پردیس 2 فنی شماره ثبت: E 2178;کتابخانه مرکزی -تالار اطلاع رسانی شماره ثبت: 56319
- تاریخ دفاع
- ۲۸ شهریور ۱۳۹۱
- دانشجو
- بهنام خدابنده لو
- استاد راهنما
- احمد خونساری
- چکیده
- چکیده
- Abstract
- Recently, statistical frameworks have been introduced as an effective technique compared to classical deterministic methods to control the process variation and provide higher flexibility in the different execution time of tasks and also to balance temperature and performance-yield of modern embedded system design. However, satisfying several objectives that are interdependent, such as meeting the deadline of the application and keeping the temperature low and controlling the performance-yield of the chip simultaneously is a challenging problem. In this thesis, we present a temperature-aware quasi-static task allocation and scheduling algorithm under process variation for hard real-time applications on MPSoCs. There are two methods for solving this problem namely stochastic method and scenario-based method and we implemented these methods. In stochastic method, first we define a new parameter as “valid frequency interval” and then we model this by a Mixed Integer Linear programming. In scenario-based method, we propose an algorithm that finds a set of schedules which maximize performance-yield and simultaneously minimize the expected value of peak temperature for periodic tasks. Proposed hierarchical framework is models the temporal correlated process variation parameters accurately to reduce the packaging variation effects and thus variation execution time of an application. To reduce the execution time of the framework, a random based weighted selection algorithm is used. The experimental results show that, using the proposed framework for 22nm technologies, significant improvements in performance-yield and peak temperature reduction can be achieved for almost all of the test cases on heterogeneous or homogenous MPSoC, which confirms the effectiveness of presented approach even though the classical methods do not find any solution.
