A novel deadline assignment strategy for a large batch of parallel tasks with soft deadlines in the cloud

Deadline assignment is to assign each subtask composing a distributed task with a local deadline such that the global deadline can be met. Today's real-time systems often need to handle hundreds or even thousands of concurrent customer (or service) requests. Therefore, deadline assignment is becoming an increasingly challenging issue with a large number of parallel and distributed subtasks. However, most conventional strategies are designed to deal with a single independent task rather than a batch of many parallel tasks in a shared resource environment such as cloud computing. To address such an issue, in this paper, instead of assigning local deadline for each subtask, we propose a novel strategy which can efficiently assign local throughput constraints for a batch of parallel tasks at any time point along the system timeline. The basis of this strategy is a novel throughput consistency model which can measure the probability of on-time completion at any given time point. The experimental results demonstrate that our strategy can achieve significant time reduction in deadline assignment and achieve the most 'consistency' between global and local deadlines compared with other representative strategies