A quantitative approach to coordinated scaling of resources in complex cloud computing workflows

Abstract: Resource scaling is widely employed in cloud computing to adapt system operation to internal (i.e., application) and external (i.e., environment) changes. We present a quantitative approach for coordinated vertical scaling of resources in cloud computing workflows, aimed at satisfying an agreed Service Level Objective (SLO) by improving the workflow e2e response time distribution. Workflows consist of IaaS services running on dedicated clusters, statically reserved before execution. Services are composed through sequence, choice/merge, and balanced split/join blocks, and have generally distributed (i.e., non-Markovian) durations possibly over bounded supports, facilitating fitting of analytical distributions from observed data. Resource allocation is performed through an efficient heuristics guided by the mean makespans of sub-workflows. The heuristics performs a top-down visit of the hierarchy of services, and it exploits an efficient compositional method to derive the response time distribution and the mean makespan of each sub-workflow. Experimental results on a workflow with high concurrency degree appear promising for feasibility and effectiveness of the approach.

Proceedings of EPEW, pp. 309-324, Springer, 2023
Stochastic ProcessesPerformance ModelsStochastic WorkflowsApplications



copy bib | save bib | save pdf | go to publisher

🏠 Home