Cascading Failure Modes in Model-as-a-Service Architectures : When Your Dependencies Think
International Journal of Scientific Research in Civil Engineering · 2023
DOI: 10.32628/ijsrce237530 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Scientific Research in Civil Engineering, 2023. Verfasst von Sheriff Adepoju. Die Arbeit nennt 23 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Following an increased move towards machine learning becoming a runtime dependency of contemporary software systems, the Model-as-a-Service (MaaS) architecture is subjected to a new category of systemic risks where failure at one component propagates across decision-making when it comes to systems. In contrast to the failure of standard microservices, failure modes in services that rely on machine learning can silently propagate throughout workflows, multiplying uncertainty, reducing the quality of decisions, and destabilizing downstream services. In this study, the authors investigated the cascading failure modes of MaaS architectures and stated that resilience should be explicitly designed based on intelligent dependencies. Based on cascading failure theory, service-oriented architectures, and critical infrastructure systems, this study conceptualizes how a combination of data inconsistencies, variability in model behavior, orchestration misalignment, and the disruptive behavior of infrastructure amplify failure. This article proposes a resilience-based architectural viewpoint that revolves around circuit breakers, fallback plans, and graceful degradation systems that are specific to machine-learning-reliant systems. Circuit breakers are re-conceptualized as resilience to erratic inference behavior and delays in decision-making; fallback strategies are re-conceptualized as goal-preserving, which does not lead to dramatic service outages but gradual decay in decision faithfulness; and graceful degradation is re-conceptualized as a controlled decline in decision faithfulness as opposed to a sudden crash in the service
Zitieren
APA
Adepoju, S. (2023). Cascading Failure Modes in Model-as-a-Service Architectures : When Your Dependencies Think. International Journal of Scientific Research in Civil Engineering. https://doi.org/10.32628/ijsrce237530
Vancouver
Adepoju S. Cascading Failure Modes in Model-as-a-Service Architectures : When Your Dependencies Think. International Journal of Scientific Research in Civil Engineering. 2023. doi:10.32628/ijsrce237530
BibTeX
@article{Adepoju2023,
author = {Adepoju, Sheriff},
title = {Cascading Failure Modes in Model-as-a-Service Architectures : When Your Dependencies Think},
journal = {International Journal of Scientific Research in Civil Engineering},
year = {2023},
doi = {10.32628/ijsrce237530}
}
EUVIMED