Tainã Coleman

Tainã Coleman, Hena Ahmed, Ravi Shende et al.

Distributed computing systems are essential for meeting the demands of modern applications, yet transitioning from single-system to distributed environments presents significant challenges. Misallocating resources in shared systems can lead to resource contention, system instability, degraded performance, priority inversion, inefficient utilization, increased latency, and environmental impact. We present BanditWare, an online recommendation system that dynamically selects the most suitable hardware for applications using a contextual multi-armed bandit algorithm. BanditWare balances exploration and exploitation, gradually refining its hardware recommendations based on observed application performance while continuing to explore potentially better options. Unlike traditional statistical and machine learning approaches that rely heavily on large historical datasets, BanditWare operates online, learning and adapting in real-time as new workloads arrive. We evaluated BanditWare on three workflow applications: Cycles (an agricultural science scientific workflow) BurnPro3D (a web-based platform for fire science) and a matrix multiplication application. Designed for seamless integration with the National Data Platform (NDP), BanditWare enables users of all experience levels to optimize resource allocation efficiently.

Tainã Coleman, Henri Casanova, Rafael Ferreira da Silva

Scientific workflow applications have become mainstream and their automated and efficient execution on large-scale compute platforms is the object of extensive research and development. For these efforts to be successful, a solid experimental methodology is needed to evaluate workflow algorithms and systems. A foundation for this methodology is the availability of realistic workflow instances. Dozens of workflow instances for a few scientific applications are available in public repositories. While these are invaluable, they are limited: workflow instances are not available for all application scales of interest. To address this limitation, previous work has developed generators of synthetic, but representative, workflow instances of arbitrary scales. These generators are popular, but implementing them is a manual, labor-intensive process that requires expert application knowledge. As a result, these generators only target a handful of applications, even though hundreds of applications use workflows in production. In this work, we present WfChef, a framework that fully automates the process of constructing a synthetic workflow generator for any scientific application. Based on an input set of workflow instances, WfChef automatically produces a synthetic workflow generator. We define and evaluate several metrics for quantifying the realism of the generated workflows. Using these metrics, we compare the realism of the workflows generated by WfChef generators to that of the workflows generated by the previously available, hand-crafted generators. We find that the WfChef generators not only require zero development effort (because it is automatically produced), but also generate workflows that are more realistic than those generated by hand-crafted generators.