[Paper Review] Resilience Design Patterns: A Structured Approach to Resilience at Extreme Scale

Saurabh Hukerikar, Christian Engelmann|QRU Quaderns de Recerca en Urbanisme|Aug 23, 2017

Radiation Effects in Electronics4 citations

TL;DR

This paper introduces a structured methodology for designing resilience in extreme-scale high-performance computing (HPC) systems using resilience design patterns—reusable, formalized solutions to common fault, error, and failure challenges. By organizing proven HPC resilience techniques into a catalog of patterns and a compositional design framework, the approach enables systematic integration, optimization, and portability of resilience solutions across hardware and software layers, improving protection coverage, performance, and power efficiency while supporting trade-off analysis at design time.

ABSTRACT

Reliability is a serious concern for future extreme-scale high-performance computing (HPC) systems. While the HPC community has developed various resilience solutions, the solution space remains fragmented. There are no formal methods and metrics to integrate the various HPC resilience techniques into composite solutions, nor are there methods to holistically evaluate the adequacy and efficacy of such solutions in terms of their protection coverage, and their performance & power efficiency characteristics. In this paper, we develop a structured approach to the design, evaluation and optimization of HPC resilience using the concept of design patterns. A design pattern is a general repeatable solution to a commonly occurring problem. We identify the problems caused by various types of faults, errors and failures in HPC systems and the techniques used to deal with these events. Each well-known solution that addresses a specific HPC resilience challenge is described in the form of a pattern. We develop a complete catalog of such resilience design patterns, which may be used as essential building blocks when designing and deploying resilience solutions. We also develop a design framework that enhances a designer's understanding the opportunities for integrating multiple patterns across layers of the system stack and the important constraints during implementation of the individual patterns. It is also useful for defining mechanisms and interfaces to coordinate flexible fault management across hardware and software components. The overall goal of this work is to establish a systematic methodology for the design and evaluation of resilience technologies in extreme-scale HPC systems that keep scientific applications running to a correct solution in a timely and cost-efficient manner despite frequent faults, errors, and failures of various types.

Motivation & Objective

• Address the fragmented and uncoordinated state of HPC resilience solutions, which lack formal integration methods and holistic evaluation metrics.
• Develop a structured, reusable framework to unify diverse resilience techniques—both application-level and system-based—into a coherent design methodology.
• Enable system architects, software developers, and application programmers to compose resilient solutions using standardized, portable patterns across evolving hardware and software stacks.
• Optimize resilience solutions by systematically analyzing trade-offs between performance, protection coverage, and power consumption.
• Provide a foundation for future resilience engineering in extreme-scale HPC by formalizing best practices into a pattern-based design framework.

Proposed method

• Identify recurring resilience problems in extreme-scale HPC systems, such as transient and hard errors due to technology scaling and near-threshold voltage operation.
• Formalize well-known HPC resilience techniques—such as checkpointing, replication, error detection, and recovery—into reusable design patterns with standardized structure and semantics.
• Organize the patterns into a layered, hierarchical catalog to reflect relationships across system stack layers and abstraction levels.
• Develop a design framework that supports composition of multiple patterns, coordination of mechanisms across hardware and software, and navigation of implementation constraints.
• Use the framework to explore design alternatives, evaluate solution efficacy, and optimize for performance, resilience coverage, and power efficiency.
• Apply the pattern-based approach to guide both new resilience solution development and reengineering of existing solutions for future architectures.

Experimental results

Research questions

• RQ1 How can resilience techniques in extreme-scale HPC systems be systematically integrated into composite, scalable solutions?
• RQ2 What formal structure and classification can unify diverse HPC resilience solutions into a coherent, reusable framework?
• RQ3 How can resilience patterns be composed across hardware and software layers to ensure flexibility, portability, and coordination?
• RQ4 What trade-offs between performance, protection coverage, and power efficiency can be systematically evaluated and optimized using this pattern-based approach?
• RQ5 To what extent can this methodology improve the design, evaluation, and evolution of resilience solutions in future HPC systems?

Key findings

• The paper establishes a comprehensive catalog of resilience design patterns that codify known solutions to common HPC fault, error, and failure problems, enabling reuse and standardization.
• The hierarchical organization of patterns allows designers to work at appropriate abstraction levels, from system architects to low-level component developers.
• The proposed design framework enables the composition of multiple patterns into complete resilience solutions while managing cross-layer dependencies and constraints.
• The pattern-based approach supports systematic exploration and optimization of resilience solutions, particularly in balancing performance, protection coverage, and power consumption.
• The methodology provides a foundation for future resilience engineering in extreme-scale HPC by formalizing best practices and enabling portable, extensible solutions.
• The framework enhances the portability and adaptability of resilience solutions across evolving hardware and software environments, including future exascale systems.