Problems with Replica Placement Using Data Dependency in Scientific Cloud Workflow

摘要

Cloud based scientific workflow has been proposed as a scalable and cost effective solution for scientific workflow execution. For such systems to work efficiently, data and replica placement should be performed intelligently. Data andreplica placement based on data dependency metric has been proposed in literature for one such popular platform, SwinDew-C. However, using only data dependency as a metric for replica placement could create problems. In this work we explore how data dependency based replica placement can lead to unexpected and adverse data placement in cloud scientific workflow systems.