Scaling Causality Analysis for Production Systems.

causality analysis;Computer Science;Engineering;Computer Science & Engineering

Chow, MichaelWenisch, Thomas F ;

University of Michigan

关键词: causality analysis;    Computer Science;    Engineering;    Computer Science & Engineering;   

Others  :  https://deepblue.lib.umich.edu/bitstream/handle/2027.42/135888/mcchow_1.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Causality analysis reveals how program values influence each other.It is important for debugging, optimizing, and understanding the execution ofprograms.This thesis scales causality analysis to production systemsconsisting of desktop and server applications as well as large-scale Internetservices.This enables developers to employ causality analysis to debug andoptimize complex, modern software systems.This thesis shows that it ispossible to scale causality analysis to both fine-grained instruction levelanalysis and analysis of Internet scale distributed systems with thousands ofdiscrete software components by developing and employing automated methods toobserve and reason about causality.First, we observe causality at a fine-grained instruction level by developingthe first taint tracking framework to support tracking millions of inputsources. We also introduce flexible taint tracking to allowfor scoping different queries and dynamic filtering of inputs, outputs, andrelationships.Next, we introduce the Mystery Machine, which uses a ``big data;;;; approach todiscover causal relationships between software components in a large-scaleInternet service. We leverage the fact that large-scale Internet servicesreceive a large number of requests in order to observe counterexamples tohypothesized causal relationships. Using discovered casual relationships, weidentify the critical path for request execution and use the critical pathanalysis to explore potential scheduling optimizations.Finally, we explore using causality to make data-quality tradeoffs inInternet services. A data-quality tradeoff is an explicit decision by a softwarecomponent to return lower-fidelity data in order to improve response time orminimize resource usage. We perform a study of data-quality tradeoffs in alarge-scale Internet service to show the pervasiveness of thesetradeoffs. We develop DQBarge, a system that enables better data-qualitytradeoffs by propagating critical information along the causal path of requestprocessing. Our evaluation shows that DQBarge helps Internet services mitigateload spikes, improve utilization of spare resources, and implement dynamiccapacity planning.