18 September 2019 in IEEE Access
In the design phase of real-time systems, it cannot be expected that the timing attributes of all tasks are completely specified and never changed. The increased computation times or shortened periods in a schedulable system often cause deadlines to be missed. In such situations, sensitivity analysis is an effective approach to provide quantitative indications for the design modification, by identifying the borderlines on parameter variations while keeping the systems schedulable. In this paper, we propose a new approach to analyze the sensitivity of the timing parameters of tasks with strict periods in multi-core real-time systems. We first analyze a schedulability condition to determine whether a task is schedulable on a given processor without changing the start times of the existing tasks. Then, following a game theory analogy, we design recursive algorithms to compute the permissible changes in the task timing parameters, by allowing each task to optimize its own start time and processor allocation. Finally, we conduct experiments with randomly generated tasks to show that our approach is more efficient than the existing solutions to solve the sensitivity problem. The proposed approach has a wide range of applications, only guiding the design of multi-core systems, but also improving the robustness of a design subject to future changes.
3 February 2020 in IEEE Access
As one of the latest market-oriented resource provisioning paradigms, cloud computing has been widely adopted by a growing number of consumers due to its powerful computing ability and storage ability. Although cloud computing can achieve effective cost reduction and convenience enhancement in the development of large-scale applications, it results in a complex cost optimization problem for data-dependent tasks represented by a workflow. All tasks in a workflow should be scheduled according to a proper strategy such that the cost is minimized and the precedence constraints and timing requirements are satisfied. In this paper, we study the cost optimization problem of deadline constrained workflows on cloud computing, and propose two list scheduling algorithms named Look-back Workflow Scheduling (LBWS) and Structure Aware Workflow Scheduling (SAWS) to solve the problem. LBWS distributes the deadline over the workflow as sub-deadlines to tasks in different levels, and schedules the tasks according to their priorities to the resources which meet their sub-deadlines and the best time-cost trade off requirements. Compared with LBWS, SAWS considers tasks allocated to the same level at a time and provisions resources with minimum cost to these tasks. Experiments on scientific workflow applications with different data and computational characteristics are conducted to show that, the proposed approaches can achieve better performance in terms of success rate and monetary cost.