Performance-Feedback Autoscaling with Budget Constraints for Cloud-based\n Workloads of Workflows

The growing popularity of workflows in the cloud domain promoted the\ndevelopment of sophisticated autoscaling policies that allow automatic\nallocation and deallocation of resources. However, many state-of-the-art\nautoscaling policies for workflows are mostly plan-based or designed for\nbatches (ensembles) of workflows. This reduces their flexibility when dealing\nwith workloads of workflows, as the workloads are often subject to\nunpredictable resource demand fluctuations. Moreover, autoscaling in clouds\nalmost always imposes budget constraints that should be satisfied. The\nbudget-aware autoscalers for workflows usually require task runtime estimates\nto be provided beforehand, which is not always possible when dealing with\nworkloads due to their dynamic nature. To address these issues, we propose a\nnovel Performance-Feedback Autoscaler (PFA) that is budget-aware and does not\nrequire task runtime estimates for its operation. Instead, it uses the\nperformance-feedback loop that monitors the average throughput on each resource\ntype. We implement PFA in the popular Apache Airflow workflow management\nsystem, and compare the performance of our autoscaler with other two\nstate-of-the-art autoscalers, and with the optimal solution obtained with the\nMixed Integer Programming approach. Our results show that PFA outperforms other\nconsidered online autoscalers, as it effectively minimizes the average job\nslowdown by up to 47% while still satisfying the budget constraints. Moreover,\nPFA shows by up to 76% lower average runtime than the competitors.\n