Cloud-based serverless computing systems, either public or privately provisioned, aim to provide the illusion of infinite resources and abstract users from details of the allocation decisions. With the goal of providing a low cost and a high QoS, the serverless computing paradigm offers opportunities that can be harnessed to attain the goals. Specifically, our strategy in this dissertation is to avoid redundant computing, in cases where independent task requests are similar to each other and for tasks that are pointless to process. We explore two main approaches to (A) reuse part of computation needed to process the services and (B) proactively pruning tasks with a low chance of success to improve the overall QoS of the system. For the first approach, we propose a mechanism to identify various types of "mergeable" tasks, which can benefit from computational reuse if they are executed together as a group. To evaluate the task merging configurations extensively, we quantify the resource-saving magnitude and then leveraging the experimental data to create a resource-saving predictor. We investigate multiple tasks merging approaches that suit different workload scenarios to determine when it is appropriate to aggregate tasks and how to allocate them so that the QoS of other tasks is minimally affected. For the second approach, we developed the mechanisms to skip tasks whose chance of completing on time is not worth pursuing by drop or defer them. We determined the minimum chance of success thresholds for tasks to pass to get scheduled and executed. We dynamically adjust such thresholds based on multiple characteristics of the arriving workload and the system's conditions. We employed approximate computing to reduce the pruning mechanism's computational overheads and ensure that the mechanism can be used practically.
翻译:无云型公共或私人提供的无云型计算机系统旨在提供无限资源和抽象用户对分配决定细节的幻觉。为了提供低成本和高QOS,无服务器计算模式为实现目标提供了可以利用的机会。具体地说,我们在这份论文中的策略是避免冗余计算,如果独立任务要求彼此相似,而且对于处理工作没有意义的任务,则避免进行冗余计算。我们探讨了两种主要办法:(A) 重新使用处理服务所需的计算部分,以及(B) 以低成功机会主动处理任务,改进系统整体的多成本类集。对于第一种办法,我们建议一种机制,以确定各种“可合并”任务,如果这些任务是作为一个小组执行,则可以从计算再利用中获益。为了广泛评估任务合并,我们量化资源节省的规模,然后利用实验数据创建资源节约的预测器。我们调查多种任务合并方法,以确保在符合不同工作量假设的情况下,能够确定总体任务的价值,从而改进系统的总体成本。我们提议一个机制的快速调整,我们决定如何将任务分配给这些任务的实际时间推向更短的周期。我们决定了其他任务。我们决定了如何完成这些任务。我们如何将时间推延延延时间,我们决定了其他任务。我们决定了这些任务。