PNNL

Abstract

Despite the extensive research and practical experience in developing quality metrics for centralized applications, there has been minimal focus on defining measures tailored specifically for distributed applications. Distributed applications pose inherently greater complexity than their centralized counterparts, primarily due to i) utilization of components operating across different network nodes and platforms, ii) the diverse characteristics ranging from data-driven to model-based and optimization-based approaches, and iii) various possible design architectures such as peer-to-peer, decentralized, and hierarchical structures. Despite these challenges, it becomes imperative to characterize these applications accurately to identify their computational requirements and assess the value/benefit of distributed design and operations. To address this, we introduce a layered framework to establish metrics across different levels. Each layer in the proposed framework focuses on distinct attributes of distributed applications, enabling users to take a bottom-up approach to identifying their specific requirements. This paper focuses on characterizing distributed application computing requirements within this framework, providing a structured methodology for effectively managing resources in distributed computing environments, and the developed metrics can allow operators to assess the advantages and disadvantages of distributed apps.