Managing shared resources in multiprocessor real-time systems can often lead to considerable schedulability sacrifice, and currently there exist no optimal multiprocessor resource sharing solutions. In addition, the choice of task mapping and priority ordering algorithms also has a direct impact on the efficiency of multiprocessor resource sharing. This thesis argues that instead of adopting a single resource sharing protocol with the traditional task mapping (e.g., the task allocation schemes that are based on utilisation only) and priority ordering (e.g., the Deadline Monotonic Priority Ordering) algorithms, the schedulability loss for managing shared resources on multiprocessors can be effectively reduced by applying a combination of appropriately chosen resource sharing protocols with new resource-oriented task allocation schemes and a new search-based priority ordering algorithm (which are independent from multiprocessor resource sharing protocols and the corresponding schedulability tests).

In this thesis, a Flexible Multiprocessor Resource Sharing (FMRS) framework is proposed that aims to provide feasible resource sharing, task allocation and priority assignment solutions to fully-partitioned systems with shared resources, where each resource is controlled by a designated locking protocol. To achieve this, the candidate resource sharing protocols for this framework are firstly determined with a new schedulability test developed to support the analysis of systems with multiple locking protocols in use. Then, besides the existing algorithms, three new resource-orientated task allocation schemes and a search-based priority ordering algorithm are developed for the FMRS framework as the task mapping and priority ordering solutions. The choices of which locking protocols, task allocation and priority ordering algorithm should be adopted to a given system are determined off-line via a genetic algorithm. As demonstrated by evaluations, the FMRS framework can facilitate multiprocessor resource sharing and has a better performance than the traditional resource control and task scheduling techniques for fully-partitioned systems.

Download Not Available

BibTex Entry

@phdthesis{PHD_Zhao,
 author = {Shuai Zhao},
 day = {1},
 month = {April},
 publisher = {University of York},
 school = {University of York},
 title = {A FIFO Spin-based Resource Control Framework for Symmetric Multiprocessing},
 url = {http://etheses.whiterose.ac.uk/21014/},
 year = {2018}
}