A Real-Time Workload Monitoring–Based Intelligent Auto-Scaling Framework for Cloud Systems

Transformer-Based Multi-Signal Predictive Autoscaling for SLA-Aware Resource Management in Kubernetes-Orchestrated Cloud-Native Environments

Kubernetes has become the dominant orchestration substrate for cloud-native applications, yet its native autoscaling mechanisms remain primarily reactive, threshold-driven, and limited in their ability to anticipate workload volatility before service-level agreement violations occur. Modern microservice systems exhibit non-linear interactions among request arrival rates, queueing delays, CPU saturation, memory pressure, network variability, pod cold-start latency, and downstream dependency bottlenecks. These characteristics make single-metric autoscaling policies insufficient for latency-sensitive workloads operating under strict service-level objectives. This paper proposes a Transformer-Based Multi-Signal Predictive Autoscaling framework for SLA-aware resource management in Kubernetes-orchestrated cloud-native environments. The proposed framework integrates heterogeneous observability signals, multi-horizon time-series forecasting, uncertainty-aware decision logic, and Kubernetes-native actuation to allocate resources before overload conditions materialize. Unlike conventional Horizontal Pod Autoscaler configurations that respond after resource utilization crosses predefined thresholds, the proposed approach forecasts near-future demand and performance risk using a Transformer encoder architecture designed to learn long-range dependencies, temporal seasonality, burst behavior, and cross-metric interactions. The framework translates predicted workload and latency risk into safe scaling actions through policy constraints that consider replica bounds, cooldown windows, pod readiness delays, cost budgets, and SLA violation probability. The paper develops the conceptual architecture, methodological workflow, evaluation metrics, and analytical discussion necessary for empirical implementation. The study argues that SLA-aware predictive autoscaling should be treated not merely as a forecasting task but as an integrated control problem involving observability quality, model calibration, decision governance, and runtime safety. The proposed model contributes to cloud resource management research by aligning deep temporal learning with Kubernetes operational semantics and by providing a structured pathway toward more reliable, efficient, and self-adaptive cloud-native platforms.

Adaptive Multi-Metric Autoscaling for Serverless Platforms

A hybrid congestion-aware autoscaling framework for serverless platforms inspired by TCP slow-start and congestion control principles is proposed, which significantly improves latency performance, scalability, and resource efficiency in serverless environments.

AutoSLO: Practical Latency SLOs on Cloud Data Warehouses -- Extended Version

Modern cloud data warehouses decouple compute from storage, making it easy for organizations to access the same underlying data with multiple compute clusters. This flexibility is often used for performance isolation among diverse workloads, so that each workload meets its latency service-level objective (SLO) more reliably. For example, interactive dashboards, ad hoc analysis, and batch jobs can each run on separate clusters. However, this dedicated-cluster approach requires each compute cluster to be continuously scaled to adapt to workload evolution, with over-provisioning wasting resources and under-provisioning risking SLO violations. We present AutoSLO, a latency-SLO-aware workload management framework for multi-cluster cloud data warehouses. AutoSLO operates across three timescales through three key components. First, a periodic Policy Tuner plans proactive cluster scaling actions and tunes configuration parameters, using simulations of history-derived workload forecasts. Second, an SLO-aware reactive Autoscaler adjusts the active cluster set when recent workload behavior deviates from the forecast. Third, an online Query Router reacts to live load when placing each query, using a concurrency-aware latency predictor to avoid SLO violations. On realistic Redbench workloads, AutoSLO successfully meets latency SLOs of varying strictness, reducing cost by a mean of 26.4% compared to the per-scenario next-best baseline. Component-level evaluations show that the Query Router and Autoscaler respectively reduce SLO violation rates by a mean of 47.8% and 93.7%, relative to their corresponding alternatives. Finally, we show that the Policy Tuner can reduce the SLO violation rate by a mean of 44.6% using a single day of workload history, and that each component is efficient given its intended operating timescale.

Regime-Aware Resource Demand Forecasting for Cloud Scheduling: When History Beats ML and When ML Matters

Efficient cluster scheduling requires reliable forecasts of resource demand, yet production workloads are heterogeneous, bursty, and strongly time-dependent. Using the Google Cluster-Usage Traces v3, we study leakage-safe prediction of three scheduling-relevant metrics: mean CPU demand, normalized memory pressure (average memory relative to assigned memory), and tail CPU demand (p95) as a burst-risk indicator. We develop a preprocessing and evaluation protocol that explicitly addresses two common threats to validity in trace-based learning: post-execution feature leakage (e.g., usage-derived fields) and identity leakage from random splits when recurring workloads appear in both training and test sets. Under time-ordered and gap-based splits, we compare gradient-boosted tree models (LightGBM) against strong history-only baselines (LastSeen and EMA) and perform a cold-start analysis by evaluating the first K occurrences of each workload entity. Results reveal a clear regime shift: for warm, recurring workloads, simple entity-history predictors achieve near-optimal accuracy and consistently outperform learned models; however, in cold-start settings where history is unavailable, LightGBM substantially improves CPU mean and tail forecasts (e.g., large gains in R2 for first-occurrence entities). These findings support a practical scheduling strategy: a hybrid, regime-aware policy that uses machine learning (ML) as a cold-start fallback and switches to lightweight history-based prediction as observations accumulate.

SLO-Aware Graph Forecasting for Intelligent Autoscaling in API and Microservice Backends

Autoscaling microservice backends is challenging due to complex inter-service dependencies, bursty workloads, and cascading latency amplification along call chains. Existing approaches either react to threshold breaches after SLO violations occur, or predict demand per service in isolation. We propose SAGF (SLO-Aware Graph Forecasting), an integrated prediction-and-control framework for proactive autoscaling. SAGF encodes six per-service metrics on a spatio-temporal call graph, predicts per-service SLO risk via GCN with temporal gating, and feeds these scores to a constrained RL controller for dependency-aware scaling. On the Online Boutique benchmark with Kubernetes and Istio, SAGF reduces SLO violations by 41% and excess overprovision by 48% on average compared to the strongest baseline across four workload patterns.

Towards Practical Latency SLOs on Cloud Data Warehouses