14 papers selected.
Low-Depth Spatial Tree Algorithms
Yves Baumann, Tal Ben-Nun, Maciej Besta, Lukas Gianinazzi et al.
TL;DR — Introduces parallel spatial-tree algorithms with provably low depth, advancing the theory of work-efficient parallel data structures for geometric workloads.
Alternative Basis Matrix Multiplication is Fast and Stable
Oded Schwartz, Sivan Toledo, Noa Vaknin, Gal Wiernik
TL;DR — Demonstrates that alternative-basis matrix multiplication achieves both practical speed and numerical stability, challenging the conventional trade-off between the two.
Wait-free Trees with Asymptotically-Efficient Range Queries
Ilya Kokorin, Victor Yudov, Vitaly Aksenov, Dan Alistarh
TL;DR — Presents the first wait-free balanced search tree supporting asymptotically optimal range queries, a long-standing open problem in concurrent data structures.
Parallel Derandomization for Coloring
Sam Coy, Artur Czumaj, Peter Davies-Peck, Gopinath Mishra
TL;DR — Develops deterministic parallel graph-coloring algorithms via derandomization, closing a key gap between randomized and deterministic complexity in this foundational problem.
HINT: Designing Cache-Efficient MPI_Alltoall using Hybrid Memory Copy Ordering and Non-Temporal Instructions
Bharath Ramesh 0005, Nick Contini, Nawras Alnaasan, Kaushik Kandadi Suresh et al.
TL;DR — Achieves substantial MPI_Alltoall bandwidth improvements by combining cache-aware copy ordering with non-temporal store instructions, directly benefiting large-scale collective communication.
An Optimized Error-controlled MPI Collective Framework Integrated with Lossy Compression
Jiajun Huang 0001, Sheng Di, Xiaodong Yu 0001, Yujia Zhai et al.
TL;DR — Integrates error-bounded lossy compression directly into MPI collectives, reducing communication volume with provable accuracy guarantees for HPC scientific applications.
Software Resource Disaggregation for HPC with Serverless Computing
Marcin Copik, Marcin Chrapek, Larissa Schmid, Alexandru Calotoiu et al.
TL;DR — Shows that serverless computing can serve as a practical resource-disaggregation layer for HPC, enabling fine-grained elasticity without sacrificing performance.
Tackling Cold Start in Serverless Computing with Multi-Level Container Reuse
Amelie Chi Zhou, Rongzheng Huang, Zhoubin Ke, Yusen Li et al.
TL;DR — Proposes a multi-level container-reuse strategy that significantly reduces cold-start latency in serverless platforms, addressing one of the main performance bottlenecks.
LightDAG: A Low-latency DAG-based BFT Consensus through Lightweight Broadcast
Xiaohai Dai, Guanxiong Wang, Jiang Xiao 0001, Zhengxuan Guo et al.
TL;DR — Redesigns DAG-based Byzantine fault-tolerant consensus to use lightweight broadcast, cutting latency while preserving safety and liveness in distributed systems.
Benchmarking and Dissecting the Nvidia Hopper GPU Architecture
Weile Luo, Ruibo Fan, Zeyu Li, Dayou Du et al.
TL;DR — Provides the first systematic microbenchmark characterization of Hopper’s new hardware features (TMA, warpgroup MMA, NVLink-4), yielding actionable insights for kernel developers.
DEFCON: Deformable Convolutions Leveraging Interval Search and GPU Texture Hardware
Malith Jayaweera, Yanyu Li, Yanzhi Wang 0001, Bin Ren 0002 et al.
TL;DR — Exploits GPU texture-cache hardware to accelerate deformable convolutions, delivering significant speedups over cuDNN-based baselines for irregular memory-access patterns.
nOS-V: Co-Executing HPC Applications Using System-Wide Task Scheduling
David Álvarez 0006, Kevin Sala, Vicenç Beltran 0001
TL;DR — Introduces a system-wide task scheduler that safely co-executes multiple HPC applications on shared hardware, improving cluster utilization without modifying application code.
Hadar: Heterogeneity-Aware Optimization-Based Online Scheduling for Deep Learning Cluster
Abeda Sultana, Fei Xu, Xu Yuan 0001, Li Chen 0019 et al.
TL;DR — Formulates deep-learning cluster scheduling as an online optimization problem that explicitly accounts for GPU heterogeneity, reducing job completion times and improving fairness.
A Parallel Partial Merge Repair Algorithm for Multi-block Failures for Erasure Storage Systems
Shuaipeng Zhang, Shiyi Li, Chentao Wu, Ruobin Wu et al.
TL;DR — Presents a parallel repair algorithm for simultaneous multi-block erasure failures that outperforms sequential recovery while reducing I/O and computational overhead.