[論文レビュー] A Formal Framework for Predicting Distributed System Performance under Faults (Extended Version)

Ziwei Zhou, Si Liu+3 more|arXiv (Cornell University)|Feb 22, 2026

Software System Performance and Reliability被引用数 0

ひとこと要約

The paper introduces PerF, a Maude-based formal framework with a fault library and model composition to predict distributed system performance under diverse faults, validated against real deployments.

ABSTRACT

Today's distributed systems operate in complex environments that inevitably involve faults and even adversarial behaviors. Predicting their performance under such environments directly from formal designs remains a longstanding challenge. We present the first formal framework that systematically enables performance prediction of distributed systems across diverse faulty scenarios. Our framework features a fault injector together with a wide range of faults, reusable as a library, and model compositions that integrate the system and the fault injector into a unified model suitable for statistical analysis of performance properties such as throughput and latency. We formalize the framework in Maude and implement it as an automated tool, PERF. Applied to representative distributed systems, PERF accurately predicts system performance under varying fault settings, with estimations from formal designs consistent with evaluations on real deployments.

研究の動機と目的

• Address the gap between formal design and performance under faults in distributed systems.
• Provide a modular, reusable fault library that can be composed with system models.
• Enable automatic generation of fault-injected models and quantitative performance analysis.
• Support a range of faults including benign and Byzantine to cover realistic environments.
• Offer a tool (PerF) that integrates with statistical model checking for end-to-end analysis.

提案手法

• Model distributed systems and faults as probabilistic rewrite theories in Maude.
• Represent faults as actors (fault handlers) that interact with system actors via messages.
• Compose system models with a fault injector to produce an integrated model.
• Use a fault-behavior priority scheme to resolve multiple applicable faults.
• Prove that the composed model preserves absence of nondeterminism (AND) for reliable statistical model checking.
• Automate fault injection, model transformation with event monitoring, and QuaTEx-based performance analysis via PVeStA.

実験結果

リサーチクエスチョン

• RQ1 How can faults be systematically integrated into formal system models for performance analysis?
• RQ2 Can a modular fault library support diverse fault types and compositions while preserving analyzability?
• RQ3 How accurately can formal models predict throughput and latency under varying fault conditions compared to real deployments?
• RQ4 What guarantees (e.g., absence of nondeterminism) are required to enable statistical model checking of fault-injected systems?

主な発見

• PerF accurately predicts performance under various faults across six distributed systems, with model-based results aligning with deployment evaluations.
• A reusable fault library covers benign and Byzantine faults and supports modular composition with systems.
• Fault prioritization ensures deterministic, analyzable interactions during fault injection.
• The framework preserves the necessary AND property, enabling end-to-end statistical model checking with QuaTEx properties.
• Experimental setup uses real deployments across CloudLab and Tencent Cloud to validate predictions under faults like message loss, delays, crashes, partitions, and equivocation.