Research Archive

Journal Article Open Access Software Engineering

Cultural Transformation in DevOps Adoption: Organizational Change Management, Psychological Safety, and the Role of Leadership in Enabling Engineering Culture Shifts

The dominant discourse in DevOps adoption research has centered on toolchain selection and process automation, while the organizational and psychological dimensions of transformation have received comparatively little rigorous attention. This paper addresses that gap through a qualitative study of DevOps cultural transformation in eight organizations, drawing on 76 interviews with practitioners ranging from individual contributors to C-suite executives, supplemented by participant observation at four of the organizations over a six-month period. Grounded in organizational change theory and psychological safety research, our analysis identifies three primary cultural transformation pathways — Top-Down Mandate, Grassroots Emergence, and Centre of Excellence Diffusion — and characterizes the conditions under which each succeeds or fails. We find that psychological safety, as operationalized by Edmondson (1999), is the single most predictive cultural variable for DevOps transformation success, outperforming technology budget, leadership commitment, and prior agile experience in our qualitative comparative analysis. Transformations that explicitly cultivated psychological safety through blameless postmortems, open deployment failure communication, and junior engineer empowerment reached self-sustaining cultural momentum 2.4 times faster than those relying on mandate-only approaches. We develop a DevOps Culture Transformation Playbook comprising 11 leadership interventions mapped to transformation phase and organizational context.

Ngozi Anozie, Sebastian Fischer, Mi-Young Choi, David Osei-Poku· Dec 2016· 341 citations

Cite View PDF

Journal Article Subscription Robotics

Simultaneous Localization and Mapping for Autonomous Mobile Robots in Dynamic Indoor Environments: Comparison of EKF-SLAM, Particle Filter SLAM, and Graph-Based SLAM Under Moving Obstacle Conditions

Simultaneous Localization and Mapping (SLAM) algorithms underpin autonomous navigation in mobile robots operating in indoor environments, yet the assumption of a static world that governs most published SLAM formulations is systematically violated in real deployment environments -- warehouses, hospitals, and offices -- where humans, vehicles, and other robots constitute dynamic obstacles. This paper presents a rigorous comparative evaluation of three SLAM algorithm families -- Extended Kalman Filter SLAM (EKF-SLAM), Rao-Blackwellized Particle Filter SLAM (RBPF-SLAM), and Graph-Based SLAM with g2o optimization -- under controlled dynamic obstacle conditions using a standardized evaluation platform comprising a TurtleBot3 robot in a repeatable 200 square meter indoor test environment with programmable dynamic obstacle injection. We parameterize dynamic obstacle density from 0 to 40 percent of navigable space and measure localization RMSE, map consistency score, computational load, and loop closure success rate as primary evaluation dimensions. Graph-Based SLAM with dynamic object masking achieves the lowest localization RMSE (4.2 cm mean) under high obstacle density conditions, while RBPF-SLAM demonstrates the fastest recovery from localization failures (mean 8.4 seconds). We introduce the Dynamic SLAM Robustness Index (DSRI) combining localization accuracy, failure recovery speed, and computational efficiency, and provide algorithm selection guidance for six common deployment scenario profiles. All evaluation code, datasets, and robot configurations are released as an open benchmarking suite.

Adaobi Okoye, Jonas Holm, Takeshi Mori, Fatima Benali· Dec 2016· 287 citations

Cite View PDF

Journal Article Subscription Cybersecurity

Adversarial Machine Learning Attacks on Network Intrusion Detection Systems: Threat Taxonomy, Evasion Techniques, and Robust Defense Architectures

Machine learning-based Network Intrusion Detection Systems (NIDS) have been widely adopted as a replacement for signature-based approaches, yet their vulnerability to adversarial perturbations -- inputs crafted to deceive the classifier while preserving malicious functionality -- represents a critical and underaddressed threat. This paper presents a comprehensive adversarial ML threat taxonomy for NIDS, categorizing attacks across four dimensions: perturbation scope (feature-space vs problem-space), attacker knowledge (white-box, grey-box, black-box), attack timing (training-time poisoning vs inference-time evasion), and adversarial goal (evasion, impersonation, denial of service against the detector). We implement and evaluate 14 adversarial attack strategies against six representative NIDS architectures -- including Random Forest, LSTM, and Transformer-based classifiers -- using the CICIDS2017 and NSL-KDD benchmark datasets supplemented by proprietary traffic captures from a financial institution. Problem-space attacks -- which modify actual network packets rather than feature vectors -- reduce NIDS detection rates by 34-71% depending on the classifier architecture. We evaluate five defense strategies: adversarial training, ensemble diversity, feature randomization, input preprocessing, and certified robustness bounds. Adversarial training combined with ensemble diversity achieves the best robustness profile but requires 3.4x the training compute of undefended baselines. We release a standardized adversarial NIDS evaluation framework to facilitate reproducible future research.

Adaobi Nwosu, Henrik Strand, Yuki Suzuki, Amina El-Amin· Oct 2016· 412 citations

Cite View PDF

Journal Article Subscription Software Engineering

Test Automation Maturity in Agile-DevOps Organizations: A Taxonomy of Testing Strategies and Their Impact on Deployment Confidence

Test automation is widely recognized as a prerequisite for continuous delivery, yet organizations adopting DevOps frequently encounter a plateau in testing effectiveness characterized by slow feedback cycles, brittle test suites, and low confidence in automated regression results. This paper presents a taxonomy of test automation strategies observed across 16 organizations at varying stages of DevOps maturity, derived from a mixed-methods study combining semi-structured interviews (n=91), pipeline artifact analysis, and a cross-sectional survey (n=376). We classify test automation approaches across four architectural layers — unit, integration, contract, and end-to-end — and evaluate their contribution to deployment confidence, defined as engineers` willingness to deploy to production without manual verification. Our analysis demonstrates that contract testing, particularly consumer-driven contract testing using tools such as Pact, delivers the highest improvement in deployment confidence per unit of maintenance cost in microservices environments. Organizations with mature contract testing practices reported 55% fewer production incidents attributable to interface regressions. We introduce the Test Automation Value Matrix (TAVM), which plots testing strategies against cost-of-maintenance and confidence contribution dimensions, enabling teams to prioritize testing investments. We also present a five-stage Test Automation Maturity Model (TAMM) and map the organizations in our study to its levels.

Amara Okafor, Jonas Lindgren, Satoshi Kobayashi, Elena Moreira· Oct 2016· 296 citations

Cite View PDF

Journal Article Subscription Cloud Computing

Infrastructure as Code: Principles, Patterns, and Pitfalls in Cloud-Native DevOps Environments

Infrastructure as Code (IaC) represents a paradigm shift in how cloud infrastructure is provisioned, managed, and evolved, yet its practical adoption is accompanied by a set of under-documented failure patterns. This paper conducts a systematic literature review of 94 peer-reviewed publications combined with a practitioner survey of 340 DevOps engineers across North America and Europe. We categorize IaC tools into three architectural families — declarative, imperative, and hybrid — and evaluate them against six quality dimensions: idempotency, modularity, testability, auditability, portability, and drift detection. Our survey reveals that 67% of teams encounter configuration drift within six months of initial deployment, and that fewer than 30% implement automated compliance checks on their IaC manifests. We introduce the concept of "infrastructure entropy" to describe the gradual degradation of alignment between declared and actual infrastructure state, and propose a set of 14 engineering practices — collectively termed the IaC Hygiene Framework — to mitigate it. Case evidence from three organizations using Terraform, Ansible, and Pulumi respectively is used to validate the framework. This research provides both theoretical grounding and practical tooling guidance for organizations pursuing robust cloud infrastructure automation.

Dmitri Volkov, Samantha Osei-Bonsu, Jiaying Wu, Carlos Mendez-Rios· Aug 2016· 278 citations

Cite View PDF

Journal Article Open Access Artificial Intelligence

Deep Reinforcement Learning for Adaptive Resource Allocation in Multi-Tenant Cloud Data Centers: Architecture, Training Regimes, and Production Evaluation

Resource allocation in multi-tenant cloud data centers has traditionally been governed by hand-crafted heuristics that fail to adapt to the non-stationary workload distributions characteristic of production environments. This paper presents DeepAlloc, a deep reinforcement learning framework for adaptive resource allocation that replaces static scheduling policies with neural policy networks trained using Proximal Policy Optimization (PPO) across simulated and real cluster environments. DeepAlloc models the allocation problem as a Markov Decision Process in which the state space encodes current cluster utilization, pending job queue characteristics, and tenant SLA parameters, while the action space encompasses CPU core assignment, memory quota setting, and network bandwidth reservation decisions. We evaluate DeepAlloc against four baseline schedulers -- FIFO, Shortest Job First, DRF, and Kubernetes default -- using both a simulator driven by 14 months of production trace data from a large cloud provider and a 200-node physical testbed. DeepAlloc achieves 23% higher cluster utilization, 41% reduction in SLA violation rate, and 18% lower mean job completion time compared to the best-performing baseline. We characterize the training stability challenges specific to cluster scheduling environments and introduce curriculum learning and action masking techniques that reduce policy collapse incidence by 87%. This work demonstrates the practical viability of deep RL as a production scheduling substrate.

Chukwuebuka Obi, Maja Lindgren, Ryusei Tanaka, Fatima Benali· Aug 2016· 388 citations

Cite View PDF

Journal Article Open Access Natural Language Processing

Sentence Embedding Architectures for Cross-Lingual Information Retrieval: Comparative Evaluation of Multilingual LSTM, Subword Segmentation, and Shared Encoder Approaches on Low-Resource Language Pairs

Cross-lingual information retrieval -- the task of retrieving documents in one language using queries formulated in another -- demands sentence representation architectures capable of projecting semantically equivalent content from different languages into aligned embedding spaces. This challenge is particularly acute for low-resource language pairs where parallel training corpora are scarce. This paper presents a systematic comparative evaluation of three cross-lingual sentence embedding architectures: multilingual LSTM encoders with shared vocabulary, subword segmentation approaches using SentencePiece with joint BPE, and shared encoder transformer architectures pretrained with cross-lingual objectives. Evaluation is conducted across 12 language pairs spanning high-resource (English-French, English-German), medium-resource (English-Swahili, English-Yoruba), and low-resource (English-Igbo, English-Hausa) settings using a standardized retrieval benchmark we construct from Wikipedia parallel corpora. Shared encoder transformers achieve the highest mean average precision across all resource levels, but exhibit a steeper performance cliff below 50,000 parallel training sentence pairs compared to subword approaches. For the Yoruba and Igbo language pairs, subword segmentation with morphologically-informed tokenization outperforms shared encoder approaches by 11.4 and 14.7 MAP points respectively due to the agglutinative morphological structure of these languages. We release the low-resource African language retrieval benchmark as an open dataset to stimulate further research in underrepresented language families.

Ngozi Chukwu, Erik Lindstrom, Yuki Inoue, Amina Diallo· Mar 2016· 302 citations

Cite View PDF

Journal Article Open Access Software Engineering

Continuous Integration and Continuous Delivery Pipelines: Empirical Evidence from Large-Scale Enterprise Adoptions

Continuous Integration and Continuous Delivery (CI/CD) have become foundational practices in modern software engineering, yet their large-scale adoption within enterprise environments remains poorly understood. This paper presents findings from a multi-case study involving twelve enterprise organizations spanning the finance, healthcare, and telecommunications sectors, each with engineering teams of over 200 developers. Through 87 semi-structured interviews, artifact analysis, and longitudinal observation over 18 months, we identify the critical success factors and systemic barriers that determine CI/CD adoption outcomes. Our findings reveal that cultural resistance, legacy system incompatibility, and inadequate test automation maturity are the three most significant impediments. We propose a five-stage CI/CD Maturity Model (CM²) that maps organizational capabilities to adoption readiness, and validate it against our case data. Organizations that progressed beyond Stage 3 of the model reported a 43% reduction in mean time to recovery (MTTR) and a 61% increase in deployment frequency within 12 months. This work contributes both a validated empirical framework and actionable guidance for practitioners navigating enterprise-scale DevOps transformation.

Lena Hartmann, Marcus J. Oduya, Priya Subramaniam, Thomas Beckett· Mar 2016· 312 citations

Cite View PDF