Analytical Performance Modeling of Non-Deterministic Cloud Workflows Using CSM

Cloud workflows are widely used for scientific applications as it divides large applications into smaller tasks that run sequentially or in parallel. Workflows can be described as Directed Acyclic Graph (DAG) or nonDirected Acyclic Graph (non-DAG). Non-DAG workflows allow iteration patterns. Iteration and choice patterns result in a probabilistic execution cost which makes it hard to estimate execution time, required resources, cost and power consumption. This paper uses Computational Structure Model (CSM) to help analyze workflows and presents all possible executions of the workflow by creating a function of a number of the flows ina workflow. This will help designers of the workflow to estimate the execution time and cost before running the workflow in the cloud.