Online Confidence-Gated LSTM-DQN for Dynamic Cloud Resource Allocation

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.

TideRL: Boosting Agentic RL Goodput with Readiness-Aware Scheduling

TideRL is presented, a readiness-aware elastic RL system with Continuous Task Batching, Resource-Aware Ref-Actor Pipelining, and Elastic Resource Scaling that improves RL training goodput and reduces per-step training time across text-only and multi-modal agentic workloads.

CAPA: A PREDICTION-DRIVEN AUTOSCALING

Testing the Divergence Stack Memory on GPGPUs: A Modular in-Field Test Strategy

RL-ICE is proposed as an innovative scheduler that can work in such a cloud continuum by leveraging a multi-cluster and hierarchical RL to satisfy both user Quality of Experience (QoE) metrics and tenant’s costs.

Scheduling Mixed RL Rollouts Beyond Prefix Locality

Modern reinforcement learning (RL) post-training pipelines for large language models (LLMs) increasingly combine rollout workloads across multiple domains and feedback paradigms. Prefix-aware routing improves inference efficiency through cache reuse and load balancing, but it does not control how heterogeneous rollout sessions compete for KV-cache capacity. When reinforcement learning with verifiable rewards (RLVR), reinforcement learning from human feedback (RLHF), and agentic rollouts share an asynchronous inference service, their distinct sequence structures, interaction patterns, and KV-residency times create substantially different serving demands. Rollout scheduling must account for this heterogeneity without distorting the workload mixture specified by the trainer. We present MISA-T, a routing-layer admission policy for mixed rollout serving. MISA-T combines adaptive session admission, workload-aware KV-capacity allocation, and residency-time-aware KV accounting. In rollout-only ablations on Step3.7 and Qwen3.6-35B-A3B, MISA-T improves rollout throughput over a sweep-tuned cache-aware vLLM Router by 53.3% and 43.6%, respectively, while maintaining high prefix-cache hit rates. In a matched 50-iteration Step3.7 experiment, it increases rollout throughput by 35.6% and reduces mean iteration time by 22.8%, while keeping the consumed workload mixture close to the trainer target and achieving comparable task scores.

BOOSTEDSOSA: Accelerated Inferencing for Low Variance Stochastic Online Scheduling

Heterogeneous scheduling in stochastic, online envi- ronments, such as high-performance computing (HPC) systems, presents a significant challenge. Stochastic Online Scheduling Accelerators (SOSAs) offer a promising solution, but their effectiveness is compromised by a reliance on runtime estimates provided by users. These estimates introduce substantial vari- ance into the scheduling process (mean MAE in hundreds of Core-Days), thereby weakening the competitiveness of Stochastic Online Scheduling algorithms as their competitive-ratio bound increases with runtime variability. To address this limitation, we introduce BOOSTEDSOSA, a dual-FPGA ML-assisted Scheduling architecture that integrates a Machine Learning predictor for expected processing times, with a novel temporal-aware training policy. The predictor estimates job runtimes using only scheduler parameters available at submission time, enabling its use in existing HPC systems. Using historical real-world HPC job data (from the Argonne Leadership Comput- ing Facility, MIT Supercloud and UIUC Blue Waters workload datasets), we show that the predictor reduces MAE by up to 63.85% compared to user runtime estimates, and the additive training policy reduces MAE by up to 71.88% compared to a static model. End-to-end, BOOSTEDSOSA achieves an average 17x speedup over an AVX-optimized software baseline and processes up to 1,711 jobs/seconds