In high contention environments, with limited number of shared resources, elimination of resource conflicts between tasks processed in parallel is required. Execution of all tasks without resource conflicts can be achieved by preparing a proper overall schedule for all of them. The effective calculation of conflict-free execution plan for tasks provides the conflictless scheduling algorithm that is dedicated to GPU massively parallel processing. The conflictless scheduling algorithm base on rapid resource conflict detection to mutual exclusion of conflicted tasks in access to global resources and is an alternative to other task synchronization methods. This article presents the performance of modern GPU in calculations of adaptive conflictless task schedule. The performance analysis also takes into account all data transfers to and from the GPU memory in various phases of the conflictless task scheduling algorithm.
New concept of conflictless task scheduling is an alternative approach to existing solutions in concurrency. Conflictless task scheduling includes data structures and algorithm that prevents occurrence of resource conflict between tasks executed in parallel. The range of applications the conflictless task scheduling includes different environments like transactions processing in database management systems, scheduling of processes or threads in operating systems or business processes management. Task scheduling without any resource conflicts is dedicated to high contention of limited resources environments and its algorithm can be implemented in modern GPU. This paper presents concept of local task scheduling without resources conflicts occurrence, discusses features of new approach and focuses on problem of task starvation. Elimination of task starvation is included in conflictless task scheduling concept, detailed explanation are contained in this paper.
JavaScript jest wyłączony w Twojej przeglądarce internetowej. Włącz go, a następnie odśwież stronę, aby móc w pełni z niej korzystać.