arXiv subjects

Zhiling Lan

Publications and source records attributed to Zhiling Lan.

At least 19 recordsLinked to original sources

arXiv2026

MARS: A Monte Carlo Tree Search-based Adaptive and Responsive Scheduler

Modern High Performance Computing systems depend on static heuristics and manual administration for job scheduling and reservation management. Deep Reinforcement Learning (DRL) has shown promising scheduling performance but requires historical training data and fixes the optimization goal at training time, forcing operators to retrain whenever priorities shift. We introduce MARS (Monte Carlo Tree Search-based Adaptive and Responsive Scheduler), a training-free HPC scheduler whose optimization goal is configurable through a reward function rather than baked into a learned model. MARS uses a lightweight discrete-event simulator to explore the future consequences of scheduling decisions within a strict time budget, adapting to the configured reward at each scheduling cycle. We evaluate MARS on year-long production workloads from two systems at Argonne Leadership Computing Facility -- 4,360-node Theta and 560-node Polaris---under two reward functions: wait-time minimization (MARS-CW) and utilization maximization (MARS-CU). Unlike DRL and heuristics, which only react to the current queue or wait for backfill to find holes, MARS exploits look-ahead to proactively drain the system and plan around future reservations, packing the system to avoid the fragmentation and utilization drop that typically precede reservation windows. MARS-CW reduces tail wait time by 64% on Theta and 43% on Polaris over the production WFP heuristic, while MARS-CU recovers utilization in the 48 hours leading into maintenance, demonstrating that MARS can target either objective via reward reconfiguration.

cs.DC↗

arXiv2026

EnergyLens: Interpretable Closed-Form Energy Models for Multimodal LLM Inference Serving

As large language models span dense, mixture-of-experts, and state-space architectures and are deployed on heterogeneous accelerators under increasingly diverse multimodal workloads, optimising inference energy has become as critical as optimizing latency and throughput. Existing approaches either treat latency as an energy proxy or rely on data-hungry black-box surrogates. Both fail under varying parallelism strategies: latency and energy optima diverge in over 20% of configurations we tested, and black-box surrogates require hundreds of profiling samples to generalize across model families and hardware. We present EnergyLens, which uses symbolic regression as a structure-discovery tool over profiling data to derive a single twelve-parameter closed-form energy model expressed in terms of system properties such as degree of parallelism, batch size, and sequence length. Unlike black-box surrogates, EnergyLens decouples tensor and pipeline parallelism contributions and separates prefill from decode energy, making its predictions physically interpretable and actionable. Fitted from as few as 50 profiling measurements, EnergyLens achieves 88.2% Top-1 configuration selection accuracy across many evaluation scenarios compared to 60.9% for the closest prior analytical baseline, matches the predictive accuracy of ensemble ML methods with 10x fewer profiling samples, and extrapolates reliably to unseen batch sizes and hardware platforms without structural modification, making it a practical, interpretable tool for energy-optimal LLM deployment.

cs.CV↗

arXiv2026

EcoShift: Performance-Aware Power Management for Power-Constrained Heterogeneous Systems

Power-constrained HPC systems increasingly run heterogeneous CPU--GPU applications under strict cluster-wide power limits. Existing cluster-wide power management policies rely on fair-share or utilization heuristics and do not capture application-specific sensitivity to CPU and GPU power caps, leading to inefficient use of reclaimed power. We present EcoShift, a performance-aware cluster-wide power management framework. EcoShift combines online performance prediction with a dynamic-programming-based allocator to distribute reclaimed power across CPU--GPU applications for maximum average performance improvement. Through emulation-based evaluation on two heterogeneous Intel CPU and NVIDIA A100/H100 GPU platforms with diverse CPU--GPU workloads, EcoShift consistently outperforms state-of-the-art policies, achieving up to 6% average performance improvement while preserving the cluster-wide power constraint.

cs.DC↗

arXiv2026

Towards Energy Efficient Co-Scheduling in HPC

Modern multi GPU HPC systems expose substantial computational capacity, yet inefficient GPU allocation often leads to wasted energy and underutilization. In practice, GPU applications exhibit heterogeneous and nonlinear scaling, making it inefficient to always use all available GPUs. We present EcoSched, an online scheduler that jointly optimizes GPU count selection and application coscheduling to improve workload level efficiency on multi GPU systems. EcoSched uses lightweight runtime profiling to estimate relative performance across GPU counts, applies a score based policy to balance energy efficiency and idle resources, and incorporates NUMA aware placement to mitigate interference. We implement EcoSched on heterogeneous CPU GPU platforms and evaluate it with diverse workloads on H100, A100, and V100 systems. EcoSched achieves up to 14.8% energy savings, 30.1% makespan improvement, and 40.4% EDP reduction over baseline schedulers, with modest performance overhead. These results show that jointly selecting GPU counts and coscheduling actions is essential for efficient multi GPU workload execution.

cs.DC↗

arXiv2025

A Real-Time Digital Twin for Adaptive Scheduling

High-performance computing (HPC) workloads are becoming increasingly diverse, exhibiting wide variability in job characteristics, yet cluster scheduling has long relied on static, heuristic-based policies. In this work we present SchedTwin, a real-time digital twin designed to adaptively guide scheduling decisions using predictive simulation. SchedTwin periodically ingests runtime events from the physical scheduler, performs rapid what-if evaluations of multiple policies using a high-fidelity discrete-event simulator, and dynamically selects the one satisfying the administrator configured optimization goal. We implement SchedTwin as an open-source software and integrate it with the production PBS scheduler. Preliminary results show that SchedTwin consistently outperforms widely used static scheduling policies, while maintaining low overhead (a few seconds per scheduling cycle). These results demonstrate that real-time digital twins offer a practical and effective path toward adaptive HPC scheduling.

cs.DC↗

arXiv2025

SMART: A Surrogate Model for Predicting Application Runtime in Dragonfly Systems

The Dragonfly network, with its high-radix and low-diameter structure, is a leading interconnect in high-performance computing. A major challenge is workload interference on shared network links. Parallel discrete event simulation (PDES) is commonly used to analyze workload interference. However, high-fidelity PDES is computationally expensive, making it impractical for large-scale or real-time scenarios. Hybrid simulation that incorporates data-driven surrogate models offers a promising alternative, especially for forecasting application runtime, a task complicated by the dynamic behavior of network traffic. We present \ourmodel, a surrogate model that combines graph neural networks (GNNs) and large language models (LLMs) to capture both spatial and temporal patterns from port level router data. \ourmodel outperforms existing statistical and machine learning baselines, enabling accurate runtime prediction and supporting efficient hybrid simulation of Dragonfly networks.

cs.LG↗

arXiv2025

Ranking Before Serving: Low-Latency LLM Serving via Pairwise Learning-to-Rank

Efficient scheduling of large language model (LLM) inference tasks is critical for achieving low latency and high throughput, a challenge that is becoming increasingly acute with the rise of reasoning-capable LLMs whose generation lengths are highly variable. Traditional strategies like First Come, First-Serve (FCFS) often suffer from Head-of-Line (HOL) blocking, where long-running tasks delay shorter ones queued behind them. In this paper, we introduce PARS, a prompt-aware LLM task scheduler that mitigates HOL blocking by approximating shortest-job-first (SJF) scheduling through pairwise ranking with a margin ranking loss. PARS effectively predicts response-length-based task ordering directly from prompts, thereby optimizing scheduling decisions with minimal overhead. In addition, it integrates seamlessly with vLLM, a state-of-the-art LLM serving system, for the research community. Extensive experiments across multiple LLM models and real-world inference use cases, including chat, math, and code generation, demonstrate that PARS significantly reduces latency by up to 15.7x compared to the vLLM default scheduler. Cross-model evaluations demonstrate that our design generalizes effectively, allowing effective scheduling across diverse LLMs without requiring model-specific retraining.

cs.LG↗

arXiv2025

Coordinated Power Management on Heterogeneous Systems

Performance prediction is essential for energy-efficient computing in heterogeneous computing systems that integrate CPUs and GPUs. However, traditional performance modeling methods often rely on exhaustive offline profiling, which becomes impractical due to the large setting space and the high cost of profiling large-scale applications. In this paper, we present OPEN, a framework consists of offline and online phases. The offline phase involves building a performance predictor and constructing an initial dense matrix. In the online phase, OPEN performs lightweight online profiling, and leverages the performance predictor with collaborative filtering to make performance prediction. We evaluate OPEN on multiple heterogeneous systems, including those equipped with A100 and A30 GPUs. Results show that OPEN achieves prediction accuracy up to 98.29\%. This demonstrates that OPEN effectively reduces profiling cost while maintaining high accuracy, making it practical for power-aware performance modeling in modern HPC environments. Overall, OPEN provides a lightweight solution for performance prediction under power constraints, enabling better runtime decisions in power-aware computing environments.

cs.DC↗

arXiv2025

Extracting Practical, Actionable Energy Insights from Supercomputer Telemetry and Logs

As supercomputers grow in size and complexity, power efficiency has become a critical challenge, particularly in understanding GPU power consumption within modern HPC workloads. This work addresses this challenge by presenting a data co-analysis approach using system data collected from the Polaris supercomputer at Argonne National Laboratory. We focus on GPU utilization and power demands, navigating the complexities of large-scale, heterogeneous datasets. Our approach, which incorporates data preprocessing, post-processing, and statistical methods, condenses the data volume by 94% while preserving essential insights. Through this analysis, we uncover key opportunities for power optimization, such as reducing high idle power costs, applying power strategies at the job-level, and aligning GPU power allocation with workload demands. Our findings provide actionable insights for energy-efficient computing and offer a practical, reproducible approach for applying existing research to optimize system performance.

cs.DC↗

arXiv2025

Leveraging LLMs to Automate Energy-Aware Refactoring of Parallel Scientific Codes

Large language models (LLMs) are increasingly used for generating parallel scientific codes, with a primary focus on generating functionally correct code. Recent work has focused on generating performant code, with an emphasis on its execution time. However, energy efficiency is now recognized as a critical objective, given the significant power demands of large-scale compute systems. This paper addresses the research question of whether LLMs can generate energy-efficient parallel scientific codes when guided by empirical execution feedback. To answer this question, we propose LASSI-EE, an automated LLM-based refactoring framework that generates energy-efficient parallel codes through a multi-stage, iterative approach integrating runtime power profiling, energy-aware prompting, self-correcting feedback loops, and an LLM-as-a-Judge agent for screening generated code. We evaluate LASSI-EE using twenty-two representative scientific benchmarks and applications on NVIDIA A100 and AMD MI100 GPUs. The results indicate an average energy reduction of 36% for MI100 and 34% for A100, across trials that produced passing energy-reducing refactorings.

cs.AI↗

arXiv2025

Exploring Uncore Frequency Scaling for Heterogeneous Computing

High-performance computing (HPC) systems are essential for scientific discovery and engineering innovation. However, their growing power demands pose significant challenges, particularly as systems scale to the exascale level. Prior uncore frequency tuning studies have primarily focused on conventional HPC workloads running on homogeneous systems. As HPC advances toward heterogeneous computing, integrating diverse GPU workloads on heterogeneous CPU-GPU systems, it is crucial to revisit and enhance uncore scaling. Our investigation reveals that uncore frequency scales down only when CPU power approaches its TDP (Thermal Design Power), an uncommon scenario in GPU-dominant applications, resulting in unnecessary power waste in modern heterogeneous computing systems. To address this, we present MAGUS, a user-transparent uncore frequency scaling runtime for heterogeneous computing. Effective uncore tuning is inherently complex, requiring dynamic detection of application execution phases that affect uncore utilization. Moreover, any robust strategy must work across a diverse range of applications, each with unique behaviors and resource requirements. Finally, an efficient runtime should introduce minimal overhead. We incorporate several key techniques in the design of MAGUS, including monitoring and predicting memory throughput, managing frequent phase transitions, and leveraging vendor-supplied power management support. We evaluate MAGUS using a diverse set of GPU benchmarks and applications across multiple heterogeneous systems with different CPU and GPU architectures. The experimental results show that MAGUS achieves up to 27% energy savings and 26% energy-delay product (EDP) reduction compared to the default settings while maintaining a performance loss below 5% and an overhead under 1%.

cs.DC↗

arXiv2025

More for Less: Integrating Capability-Predominant and Capacity-Predominant Computing

Capability jobs (e.g., large, long-running tasks) and capacity jobs (e.g., small, short-running tasks) are two common types of workloads in high-performance computing (HPC). Different HPC systems are typically deployed to handle distinct computing workloads. For example, Theta at the Argonne Leadership Computing Facility (ALCF) primarily serves capability jobs, while Cori at the National Energy Research Scientific Computing Center (NERSC) predominantly handles capacity workloads. However, this segregation often leads to inefficient resource utilization and higher costs due to the need for operating separate computing platforms. This work examines what-if scenarios for integrating siloed platforms. Specifically, we collect and characterize two real workloads from production systems at DOE laboratories, representing capabilitypredominant and capacity-predominant computing, respectively. We investigate two approaches to unification. Workload fusion explores how efficiently resources are utilized when a unified system accommodates diverse workloads, whereas workload injection identifies opportunities to enhance resource utilization on capability computing systems by leveraging capacity jobs. Finally, through extensive trace-based, event-driven simulations, we explore the potential benefits of co-scheduling both types of jobs on a unified system to enhance resource utilization and reduce costs, offering new insights for future research in unified computing.

cs.DC↗

arXiv2024

LASSI: An LLM-based Automated Self-Correcting Pipeline for Translating Parallel Scientific Codes

This paper addresses the problem of providing a novel approach to sourcing significant training data for LLMs focused on science and engineering. In particular, a crucial challenge is sourcing parallel scientific codes in the ranges of millions to billions of codes. To tackle this problem, we propose an automated pipeline framework called LASSI, designed to translate between parallel programming languages by bootstrapping existing closed- or open-source LLMs. LASSI incorporates autonomous enhancement through self-correcting loops where errors encountered during the compilation and execution of generated code are fed back to the LLM through guided prompting for debugging and refactoring. We highlight the bi-directional translation of existing GPU benchmarks between OpenMP target offload and CUDA to validate LASSI. The results of evaluating LASSI with different application codes across four LLMs demonstrate the effectiveness of LASSI for generating executable parallel codes, with 80% of OpenMP to CUDA translations and 85% of CUDA to OpenMP translations producing the expected output. We also observe approximately 78% of OpenMP to CUDA translations and 62% of CUDA to OpenMP translations execute within 10% of or at a faster runtime than the original benchmark code in the same language.

cs.SE↗

arXiv2024

Modeling and Analysis of Application Interference on Dragonfly+

Dragonfly class of networks are considered as promising interconnects for next-generation supercomputers. While Dragonfly+ networks offer more path diversity than the original Dragonfly design, they are still prone to performance variability due to their hierarchical architecture and resource sharing design. Event-driven network simulators are indispensable tools for navigating complex system design. In this study, we quantitatively evaluate a variety of application communication interactions on a 3,456-node Dragonfly+ system by using the CODES toolkit. This study looks at the impact of communication interference from a user's perspective. Specifically, for a given application submitted by a user, we examine how this application will behave with the existing workload running in the system under different job placement policies. Our simulation study considers hundreds of experiment configurations including four target applications with representative communication patterns under a variety of network traffic conditions. Our study shows that intra-job interference can cause severe performance degradation for communication-intensive applications. Inter-job interference can generally be reduced for applications with one-to-one or one-to-many communication patterns through job isolation. Application with one-to-all communication pattern is resilient to network interference.

cs.NI↗

arXiv2024

Union: An Automatic Workload Manager for Accelerating Network Simulation

With the rapid growth of the machine learning applications, the workloads of future HPC systems are anticipated to be a mix of scientific simulation, big data analytics, and machine learning applications. Simulation is a great research vehicle to understand the performance implications of co-running scientific applications with big data and machine learning workloads on large-scale systems. In this paper, we present Union, a workload manager that provides an automatic framework to facilitate hybrid workload simulation in CODES. Furthermore, we use Union, along with CODES, to investigate various hybrid workloads composed of traditional simulation applications and emerging learning applications on two dragonfly systems. The experiment results show that both message latency and communication time are important performance metrics to evaluate network interference. Network interference on HPC applications is more reflected by the message latency variation, whereas ML application performance depends more on the communication time.

cs.DC↗

arXiv2024

MRSch: Multi-Resource Scheduling for HPC

Emerging workloads in high-performance computing (HPC) are embracing significant changes, such as having diverse resource requirements instead of being CPU-centric. This advancement forces cluster schedulers to consider multiple schedulable resources during decision-making. Existing scheduling studies rely on heuristic or optimization methods, which are limited by an inability to adapt to new scenarios for ensuring long-term scheduling performance. We present an intelligent scheduling agent named MRSch for multi-resource scheduling in HPC that leverages direct future prediction (DFP), an advanced multi-objective reinforcement learning algorithm. While DFP demonstrated outstanding performance in a gaming competition, it has not been previously explored in the context of HPC scheduling. Several key techniques are developed in this study to tackle the challenges involved in multi-resource scheduling. These techniques enable MRSch to learn an appropriate scheduling policy automatically and dynamically adapt its policy in response to workload changes via dynamic resource prioritizing. We compare MRSch with existing scheduling methods through extensive tracebase simulations. Our results demonstrate that MRSch improves scheduling performance by up to 48% compared to the existing scheduling methods.

cs.DC↗

arXiv2024

Interpretable Modeling of Deep Reinforcement Learning Driven Scheduling

In the field of high-performance computing (HPC), there has been recent exploration into the use of deep reinforcement learning for cluster scheduling (DRL scheduling), which has demonstrated promising outcomes. However, a significant challenge arises from the lack of interpretability in deep neural networks (DNN), rendering them as black-box models to system managers. This lack of model interpretability hinders the practical deployment of DRL scheduling. In this work, we present a framework called IRL (Interpretable Reinforcement Learning) to address the issue of interpretability of DRL scheduling. The core idea is to interpret DNN (i.e., the DRL policy) as a decision tree by utilizing imitation learning. Unlike DNN, decision tree models are non-parametric and easily comprehensible to humans. To extract an effective and efficient decision tree, IRL incorporates the Dataset Aggregation (DAgger) algorithm and introduces the notion of critical state to prune the derived decision tree. Through trace-based experiments, we demonstrate that IRL is capable of converting a black-box DNN policy into an interpretable rulebased decision tree while maintaining comparable scheduling performance. Additionally, IRL can contribute to the setting of rewards in DRL scheduling.

cs.LG↗

arXiv2024

Study of Workload Interference with Intelligent Routing on Dragonfly

Dragonfly interconnect is a crucial network technology for supercomputers. To support exascale systems, network resources are shared such that links and routers are not dedicated to any node pair. While link utilization is increased, workload performance is often offset by network contention. Recently, intelligent routing built on reinforcement learning demonstrates higher network throughput with lower packet latency. However, its effectiveness in reducing workload interference is unknown. In this work, we present extensive network simulations to study multi-workload contention under different routing mechanisms, intelligent routing and adaptive routing, on a large-scale Dragonfly system. We develop an enhanced network simulation toolkit, along with a suite of workloads with distinctive communication patterns. We also present two metrics to characterize application communication intensity. Our analysis focuses on examining how different workloads interfere with each other under different routing mechanisms by inspecting both application-level and network-level metrics. Several key insights are made from the analysis.

cs.NI↗