arXiv subjects

Torsten Hoefler

Publications and source records attributed to Torsten Hoefler.

At least 55 records · Page 3Linked to original sources

arXiv2026

Spritz: Path-Aware Load Balancing in Low-Diameter Networks

Low-diameter topologies such as Dragonfly and Slim Fly are increasingly adopted in HPC and datacenter networks, yet existing load balancing techniques either rely on proprietary in-network mechanisms or fail to utilize the full path diversity of these topologies. We introduce Spritz, a flexible sender-based load balancing framework that shifts adaptive topology-aware routing to the endpoints using only standard Ethernet features. We propose two algorithms, Spritz-Scout and Spritz-Spray that, respectively, explore and adaptively cache efficient paths using ECN, packet trimming, and timeout feedback. Through simulation on Dragonfly and Slim Fly topologies with over 1000 endpoints, Spritz outperforms ECMP, UGAL-L, and prior sender-based approaches by up to 1.8x in flow completion time under AI training and datacenter workloads, while offering robust failover with performance improvements of up to 25.4x under link failures, all without additional hardware support. Spritz enables datacenter-scale, commodity Ethernet networks to efficiently leverage low-diameter topologies, offering unified routing and load balancing for the Ultra Ethernet era.

cs.NI↗

arXiv2026

SMaRTT: Sender-based Marked Rapidly-adapting Trimmed & Timed Transport

With the rapid growth of artificial intelligence (AI) workloads in datacenters, the Ultra Ethernet Consortium (UEC) has defined a new high-performance transport layer to deliver the required performance at scale. A core component of this new standard is the Network Signal-based Congestion Control (NSCC) algorithm. This paper presents SMaRTT, the algorithm that forms the basis of the UEC NSCC specification. SMaRTT is a sender-based congestion control algorithm that systematically combines delay, Explicit Congestion Notification (ECN), and optional packet trimming into a cohesive state machine for fast, fair and precise window adjustments with seamless multipath support. At its core lies the novel QuickAdapt algorithm that accurately estimates and rapidly adapts to available capacity. Our evaluation shows that SMaRTT outperforms existing datacenter congestion control algorithms like Swift, RoCE, and MPRDMA by up to 50% and provides superior fairness, validating the design choices made in the UEC standard.

cs.NI↗

arXiv2026

Hazel: Secure and Efficient Disaggregated Storage

Disaggregated storage with NVMe-over-Fabrics (NVMe-oF) has emerged as the standard solution in modern supercomputers and data center clusters, achieving superior performance, resource utilization, and power efficiency. Simultaneously, confidential computing (CC) is becoming the de facto security paradigm, enforcing stronger isolation and protection for sensitive workloads. However, securing state-of-the-art storage with traditional CC methods struggles to scale and compromises performance or security. To address these issues, we introduce Hazel, a storage management system that extends the NVMe-oF protocol capabilities and adheres to the CC threat model, providing confidentiality, integrity, and freshness guarantees. Hazel offers an appropriate control path with novel concepts such as counter-leasing. Hazel also optimizes data path performance by leveraging NVMe metadata and introducing a new disaggregated Hazel Merkle Tree (HMT), all while remaining compatible with NVMe-oF. For additional efficiency, Hazel also supports offloading to CC-capable smart NIC accelerators. We prototype Hazel on an NVIDIA BlueField-3 and demonstrate that it can achieve as little as 1-2% performance degradation for synthetic patterns, AI training, IO500, and YCSB.

cs.CR↗

arXiv2026

REPS: Recycled Entropy Packet Spraying for Adaptive Load Balancing and Failure Mitigation

Next-generation datacenters require highly efficient network load balancing to manage the growing scale of artificial intelligence (AI) training and general datacenter traffic. However, existing Ethernet-based solutions, such as Equal Cost Multi-Path (ECMP) and oblivious packet spraying (OPS), struggle to maintain high network utilization due to both increasing traffic demands and the expanding scale of datacenter topologies, which also exacerbate network failures. To address these limitations, we propose REPS, a lightweight decentralized per-packet adaptive load balancing algorithm designed to optimize network utilization while ensuring rapid recovery from link failures. REPS adapts to network conditions by caching good-performing paths. In case of a network failure, REPS re-routes traffic away from it in less than 100 microseconds. REPS is designed to be deployed with next-generation out-of-order transports, such as Ultra Ethernet, and uses less than 25 bytes of per-connection state regardless of the topology size. We extensively evaluate REPS in large-scale simulations and FPGA-based NICs.

cs.NI↗

arXiv2026

In-Network Collective Operations: Game Changer or Challenge for AI Workloads?

This paper summarizes the opportunities of in-network collective operations (INC) for accelerated collective operations in AI workloads. We provide sufficient detail to make this important field accessible to non-experts in AI or networking, fostering a connection between these communities. Consider two types of INC: Edge-INC, where the system is implemented at the node level, and Core-INC, where the system is embedded within network switches. We outline the potential performance benefits as well as six key obstacles in the context of both Edge-INC and Core-INC that may hinder their adoption. Finally, we present a set of predictions for the future development and application of INC.

cs.NI↗

arXiv2026

MLIR-Forge: A Modular Framework for Language Smiths

Optimizing compilers are essential for the efficient and correct execution of software across various scientific fields. Domain-specific languages (DSL) typically use higher level intermediate representations (IR) in their compiler pipelines for domain-specific optimizations. As these IRs add to complexity, it is crucial to test them thoroughly. Random program generators have proven to be an effective tool to test compilers through differential and fuzz testing. However, developing specialized program generators for compiler IRs is not straightforward and demands considerable resources. We introduce MLIR-Forge, a novel random program generator framework that leverages the flexibility of MLIR, aiming to simplify the creation of specialized program generators. MLIR-Forge achieves this by splitting the generation process into fundamental building blocks that are language specific, and reusable program creation logic that constructs random programs from these building blocks. This hides complexity and furthermore, even the language specific components can be defined using a set of common tools. We demonstrate MLIR-Forge's capabilities by generating MLIR with built-in dialects, WebAssembly, and a data-centric program representation, DaCe -- requiring less than a week of development time in total for each of them. Using the generated programs we conduct differential testing and find 9 MLIR, 15 WebAssembly, and 774 DaCe groups of bugs with the corresponding program generators, after running them until the rate of new bugs stagnates.

cs.PL↗

arXiv2026

Practical Challenges in Executing Shor's Algorithm on Existing Quantum Platforms

Quantum computers pose a fundamental threat to widely deployed public-key cryptosystems, such as RSA and ECC, by enabling efficient integer factorization using Shor's algorithm. Theoretical resource estimates suggest that 2048-bit RSA keys could be broken using Shor's algorithm with fewer than a million noisy qubits. Although such machines do not yet exist, the availability of smaller, cloud-accessible quantum processors and open-source implementations of Shor's algorithm raises the question of what key sizes can realistically be factored with today's platforms. In this work, we experimentally investigate Shor's algorithm on several cloud-based quantum computers using publicly available implementations. Our results reveal a substantial gap between the capabilities of current quantum hardware and the requirements for factoring cryptographically relevant integers. In particular, we observe that circuit constructions still need to be highly specific for each modulus, and that machine fidelities are unstable, with high and fluctuating error rates.

quant-ph↗

arXiv2025

EDAN: Towards Understanding Memory Parallelism and Latency Sensitivity in HPC

Resource disaggregation is a promising technique for improving the efficiency of large-scale computing systems. However, this comes at the cost of increased memory access latency due to the need to rely on the network fabric to transfer data between remote nodes. As such, it is crucial to ascertain an application's memory latency sensitivity to minimize the overall performance impact. Existing tools for measuring memory latency sensitivity often rely on custom ad-hoc hardware or cycle-accurate simulators, which can be inflexible and time-consuming. To address this, we present EDAN (Execution DAG Analyzer), a novel performance analysis tool that leverages an application's runtime instruction trace to generate its corresponding execution DAG. This approach allows us to estimate the latency sensitivity of sequential programs and investigate the impact of different hardware configurations. EDAN not only provides us with the capability of calculating the theoretical bounds for performance metrics, but it also helps us gain insight into the memory-level parallelism inherent to HPC applications. We apply EDAN to applications and benchmarks such as PolyBench, HPCG, and LULESH to unveil the characteristics of their intrinsic memory-level parallelism and latency sensitivity.

cs.PF↗

arXiv2025

Epidemiology of Large Language Models: A Benchmark for Observational Distribution Knowledge

Artificial intelligence (AI) systems hold great promise for advancing various scientific disciplines, and are increasingly used in real-world applications. Despite their remarkable progress, further capabilities are expected in order to achieve more general types of intelligence. A critical distinction in this context is between factual knowledge, which can be evaluated against true or false answers (e.g., "what is the capital of England?"), and probabilistic knowledge, reflecting probabilistic properties of the real world (e.g., "what is the sex of a computer science graduate in the US?"). In this paper, our goal is to build a benchmark for understanding the capabilities of LLMs in terms of knowledge of probability distributions describing the real world. Given that LLMs are trained on vast amounts of text, it may be plausible that they internalize aspects of these distributions. Indeed, LLMs are touted as powerful universal approximators of real-world distributions. At the same time, classical results in statistics, known as curse of dimensionality, highlight fundamental challenges in learning distributions in high dimensions, challenging the notion of universal distributional learning. In this work, we develop the first benchmark to directly test this hypothesis, evaluating whether LLMs have access to empirical distributions describing real-world populations across domains such as economics, health, education, and social behavior. Our results demonstrate that LLMs perform poorly overall, and do not seem to internalize real-world statistics naturally. When interpreted in the context of Pearl's Causal Hierarchy (PCH), our benchmark demonstrates that language models do not contain knowledge on observational distributions (Layer 1 of PCH), and thus the Causal Hierarchy Theorem implies that interventional (Layer 2) and counterfactual (Layer 3) knowledge of these models is also limited.

cs.AI↗

arXiv2025

Apertus: Democratizing Open and Compliant LLMs for Global Language Environments

We present Apertus, a fully open suite of large language models (LLMs) designed to address two systemic shortcomings in today's open model ecosystem: data compliance and multilingual representation. Unlike many prior models that release weights without reproducible data pipelines or regard for content-owner rights, Apertus models are pretrained exclusively on openly available data, retroactively respecting `robots.txt` exclusions and filtering for non-permissive, toxic, and personally identifiable content. To mitigate risks of memorization, we adopt the Goldfish objective during pretraining, strongly suppressing verbatim recall of data while retaining downstream task performance. The Apertus models also expand multilingual coverage, training on 15T tokens from over 1800 languages, with ~40% of pretraining data allocated to non-English content. Released at 8B and 70B scales, Apertus approaches state-of-the-art results among fully open models on multilingual benchmarks, rivalling or surpassing open-weight counterparts. Beyond model weights, we release all scientific artifacts from our development cycle with a permissive license, including data preparation scripts, checkpoints, evaluation suites, and training code, enabling transparent audit and extension.

cs.CL↗

arXiv2025

Core Hours and Carbon Credits: Incentivizing Sustainability in HPC

Realizing a shared responsibility between providers and consumers is critical to manage the sustainability of HPC. However, while cost may motivate efficiency improvements by infrastructure operators, broader progress is impeded by a lack of user incentives. We conduct a survey of HPC users that reveals fewer than 30 percent are aware of their energy consumption, and that energy efficiency is among users' lowest priority concerns. One explanation is that existing pricing models may encourage users to prioritize performance over energy efficiency. We propose two transparent multi-resource pricing schemes, Energy- and Carbon-Based Accounting, that seek to change this paradigm by incentivizing more efficient user behavior. These two schemes charge for computations based on their energy consumption or carbon footprint, respectively, rewarding users who leverage efficient hardware and software. We evaluate these two pricing schemes via simulation, in a prototype, and a user study.

cs.DC↗

arXiv2025

Demystifying Higher-Order Graph Neural Networks

Higher-order graph neural networks (HOGNNs) and the related architectures from Topological Deep Learning are an important class of GNN models that harness polyadic relations between vertices beyond plain edges. They have been used to eliminate issues such as over-smoothing or over-squashing, to significantly enhance the accuracy of GNN predictions, to improve the expressiveness of GNN architectures, and for numerous other goals. A plethora of HOGNN models have been introduced, and they come with diverse neural architectures, and even with different notions of what the "higher-order" means. This richness makes it very challenging to appropriately analyze and compare HOGNN models, and to decide in what scenario to use specific ones. To alleviate this, we first design an in-depth taxonomy and a blueprint for HOGNNs. This facilitates designing models that maximize performance. Then, we use our taxonomy to analyze and compare the available HOGNN models. The outcomes of our analysis are synthesized in a set of insights that help to select the most beneficial GNN model in a given scenario, and a comprehensive list of challenges and opportunities for further research into more powerful HOGNNs.

cs.LG↗

arXiv2025

Flowcut Switching: High-Performance Adaptive Routing with In-Order Delivery Guarantees

Network latency severely impacts the performance of applications running on supercomputers. Adaptive routing algorithms route packets over different available paths to reduce latency and improve network utilization. However, if a switch routes packets belonging to the same network flow on different paths, they might arrive at the destination out-of-order due to differences in the latency of these paths. For some transport protocols like TCP, QUIC, and RoCE, out-of-order (OOO) packets might cause large performance drops or significantly increase CPU utilization. In this work, we propose flowcut switching, a new adaptive routing algorithm that provides high-performance in-order packet delivery. Differently from existing solutions like flowlet switching, which are based on the assumption of bursty traffic and that might still reorder packets, flowcut switching guarantees in-order delivery under any network conditions, and is effective also for non-bursty traffic, as it is often the case for RDMA.

cs.NI↗

arXiv2025

Bine Trees: Enhancing Collective Operations by Optimizing Communication Locality

Communication locality plays a key role in the performance of collective operations on large HPC systems, especially on oversubscribed networks where groups of nodes are fully connected internally but sparsely linked through global connections. We present Bine (binomial negabinary) trees, a family of collective algorithms that improve communication locality. Bine trees maintain the generality of binomial trees and butterflies while cutting global-link traffic by up to 33%. We implement eight Bine-based collectives and evaluate them on four large-scale supercomputers with Dragonfly, Dragonfly+, oversubscribed fat-tree, and torus topologies, achieving up to 5x speedups and consistent reductions in global-link traffic across different vector sizes and node counts.

cs.DC↗

arXiv2025

Inductive Loop Analysis for Practical HPC Application Optimization

Scientific computing applications heavily rely on multi-level loop nests operating on multidimensional arrays. This presents multiple optimization opportunities from exploiting parallelism to reducing data movement through prefetching and improved register usage. HPC frameworks often delegate fine-grained data movement optimization to compilers, but their low-level representations hamper analysis of common patterns, such as strided data accesses and loop-carried dependencies. In this paper, we introduce symbolic, inductive loop optimization (SILO), a novel technique that models data accesses and dependencies as functions of loop nest strides. This abstraction enables the automatic parallelization of sequentially-dependent loops, as well as data movement optimizations including software prefetching and pointer incrementation to reduce register spills. We demonstrate SILO on fundamental kernels from scientific applications with a focus on atmospheric models and numerical solvers, achieving up to 12$\times$ speedup over the state of the art.

cs.DC↗

arXiv2025

Computing the Full Earth System at 1 km Resolution

We present the first-ever global simulation of the full Earth system at 1.25 km grid spacing, achieving highest time compression with an unseen number of degrees of freedom. Our model captures the flow of energy, water, and carbon through key components of the Earth system: atmosphere, ocean, and land. To achieve this landmark simulation, we harness the power of 8192 GPUs on Alps and 20480 GPUs on JUPITER, two of the world's largest GH200 superchip installations. We use both the Grace CPUs and Hopper GPUs by carefully balancing Earth's components in a heterogeneous setup and optimizing acceleration techniques available in ICON's codebase. We show how separation of concerns can reduce the code complexity by half while increasing performance and portability. Our achieved time compression of 145.7 simulated days per day enables long studies including full interactions in the Earth system and even outperforms earlier atmosphere-only simulations at a similar resolution.

physics.ao-ph↗

arXiv2025

VEIL: Reading Control Flow Graphs Like Code

Control flow graphs (CFGs) are essential tools for understanding program behavior, yet the size of real-world CFGs makes them difficult to interpret. With thousands of nodes and edges, sophisticated graph drawing algorithms are required to present them on screens in ways that make them readable and understandable. However, being designed for general graphs, these algorithms frequently break the natural flow of execution, placing later instructions before earlier ones and obscuring critical program structures. In this paper, we introduce a set of criteria specifically tailored for CFG visualization, focusing on preserving execution order and making complex structures easier to follow. Building on these criteria, we present VEIL, a new layout algorithm that uses dominator analysis to produce clearer, more intuitive CFG layouts. Through a study of CFGs from real-world applications, we show how our method improves readability and provides improved layout performance compared to state of the art graph drawing techniques.

cs.HC↗

arXiv2025

HALO: Hadamard-Assisted Lower-Precision Optimization for LLMs

Quantized training of Large Language Models (LLMs) remains an open challenge, as maintaining accuracy while performing all matrix multiplications in low precision has proven difficult. This is particularly the case when fine-tuning pre-trained models, which can have large weight and activation outlier values that make lower-precision optimization difficult. To address this, we present HALO, a novel quantization-aware training approach for Transformers that enables accurate and efficient low-precision training by combining 1) strategic placement of Hadamard rotations in both forward and backward passes, which mitigate outliers, 2) high-performance kernel support, and 3) FSDP integration for low-precision communication. Our approach ensures that all large matrix multiplications during the forward and backward passes are executed in lower precision. Applied to LLAMA-family models, HALO achieves near-full-precision-equivalent results during fine-tuning on various tasks, while delivering up to 1.41x end-to-end speedup for full fine-tuning on RTX 4090 GPUs. HALO efficiently supports both standard and parameterefficient fine-tuning (PEFT). Our results demonstrate the first practical approach to fully quantized LLM fine-tuning that maintains accuracy in 8-bit precision, while delivering performance benefits. Code is available at https://github.com/IST-DASLab/HALO.

cs.LG↗