English
Related papers

Related papers: Flare: Flexible In-Network Allreduce

200 papers

The allreduce operation is an essential building block for many distributed applications, ranging from the training of deep learning models to scientific computing. In an allreduce operation, data from multiple hosts is aggregated together…

Distributed, Parallel, and Cluster Computing · Computer Science 2023-09-29 Daniele De Sensi , Edgar Costa Molero , Salvatore Di Girolamo , Laurent Vanbever , Torsten Hoefler

The allreduce collective operation accounts for a significant fraction of the runtime of workloads running on distributed systems. One factor determining its performance is the distance between communicating nodes, especially on networks…

Distributed, Parallel, and Cluster Computing · Computer Science 2024-03-06 Daniele De Sensi , Tommaso Bonato , David Saam , Torsten Hoefler

Many large datasets exhibit power-law statistics: The web graph, social networks, text data, click through data etc. Their adjacency graphs are termed natural graphs, and are known to be difficult to partition. As a consequence most…

Distributed, Parallel, and Cluster Computing · Computer Science 2013-12-12 Huasha Zhao , John Canny

The reduce-scatter collective operation in which $p$ processors in a network of processors collectively reduce $p$ input vectors into a result vector that is partitioned over the processors is important both in its own right and as building…

Distributed, Parallel, and Cluster Computing · Computer Science 2025-02-14 Jesper Larsson Träff

In-network computing techniques, exemplified by NVLink SHARP (NVLS), offer a promising approach to addressing the communication bottlenecks in LLM inference by offloading collective operations such as All-Reduce to switches. However, the…

Hardware Architecture · Computer Science 2026-04-09 Aojie Jiang , Kang Zhu , Zhiheng Zhang , Zhengxu Su , Juntao Liu , Yuan Du , Li Du

The capacity of offloading data and control tasks to the network is becoming increasingly important, especially if we consider the faster growth of network speed when compared to CPU frequencies. In-network compute alleviates the host CPU…

Networking and Internet Architecture · Computer Science 2021-06-02 Salvatore Di Girolamo , Andreas Kurth , Alexandru Calotoiu , Thomas Benz , Timo Schneider , Jakub Beránek , Luca Benini , Torsten Hoefler

Optimizing communication performance is imperative for large-scale computing because communication overheads limit the strong scalability of parallel applications. Today's network cards contain rather powerful processors optimized for data…

Distributed, Parallel, and Cluster Computing · Computer Science 2017-10-20 Torsten Hoefler , Salvatore Di Girolamo , Konstantin Taranov , Ryan E. Grant , Ron Brightwell

Collective communications, namely the patterns allgatherv, reduce_scatter, and allreduce in message-passing systems are optimised based on measurements at the installation time of the library. The algorithms used are set up in an…

Distributed, Parallel, and Cluster Computing · Computer Science 2020-06-24 Andreas Jocksch , Noe Ohana , Emmanuel Lanti , Vasileios Karakasis , Laurent Villard

Optical circuit-switched networks have emerged as an appealing alternative to electrical fabrics as they can reconfigure the network topology at runtime, reducing communication cost and improving bandwidth utilization. Yet exploiting…

Networking and Internet Architecture · Computer Science 2026-05-14 Anton Juerss , Stefan Schmid

In grid networks, distributed resources are interconnected by wide area network to support compute and data-intensive applications, which require reliable and efficient transfer of gigabits (even terabits) of data. Different from…

Networking and Internet Architecture · Computer Science 2016-08-16 Bin Bin Chen , Pascale Primet

The \texttt{MPI\_Allreduce} collective operation is a core kernel of many parallel codebases, particularly for reductions over a single value per process. The commonly used allreduce recursive-doubling algorithm obtains the lower bound…

Distributed, Parallel, and Cluster Computing · Computer Science 2019-10-23 Amanda Bienz , Luke N. Olson , William D. Gropp

To increase the scalability of Software Defined Networks (SDNs), flow aggregation schemes have been proposed to merge multiple mouse flows into an elephant aggregated flow for traffic engineering. In this paper, we first notice that the…

Networking and Internet Architecture · Computer Science 2017-08-16 Jian-Jhih Kuo , Chih-Hang Wang , Cheng-Da Tsai , De-Nian Yang , Wen-Tsuen Chen

AllReduce is a technique in distributed computing which saw use in many critical applications of deep learning. Existing methods of AllReduce scheduling oftentimes lack flexibility due to being topology-specific or relying on extensive…

Networking and Internet Architecture · Computer Science 2025-03-28 Yufan Wei , Mickel Liu , Wenfei Wu

Distributed computing frameworks such as MapReduce are often used to process large computational jobs. They operate by partitioning each job into smaller tasks executed on different servers. The servers also need to exchange intermediate…

Distributed, Parallel, and Cluster Computing · Computer Science 2020-04-20 Konstantinos Konstantinidis , Aditya Ramamoorthy

We present NetReduce, a novel RDMA-compatible in-network reduction architecture to accelerate distributed DNN training. Compared to existing designs, NetReduce maintains a reliable connection between end-hosts in the Ethernet and does not…

Networking and Internet Architecture · Computer Science 2020-09-22 Shuo Liu , Qiaoling Wang , Junyi Zhang , Qinliang Lin , Yao Liu , Meng Xu , Ray C. C. Chueng , Jianfei He

MapReduce is a commonly used framework for executing data-intensive jobs on distributed server clusters. We introduce a variant implementation of MapReduce, namely "Coded MapReduce", to substantially reduce the inter-server communication…

Distributed, Parallel, and Cluster Computing · Computer Science 2015-12-08 Songze Li , Mohammad Ali Maddah-Ali , A. Salman Avestimehr

Collective operations are common features of parallel programming models that are frequently used in High-Performance (HPC) and machine/ deep learning (ML/ DL) applications. In strong scaling scenarios, collective operations can negatively…

Distributed, Parallel, and Cluster Computing · Computer Science 2022-04-01 Roman Iakymchuk , Amandio Faustino , Andrew Emerson , Joao Barreto , Valeria Bartsch , Rodrigo Rodrigues , Jose C. Monteiro

The advent of switches with programmable dataplanes has enabled the rapid development of new network functionality, as well as providing a platform for acceleration of a broad range of application-level functionality. However, existing…

Networking and Internet Architecture · Computer Science 2021-12-14 Yifan Yuan , Omar Alama , Amedeo Sapio , Jiawei Fei , Jacob Nelson , Dan R. K. Ports , Marco Canini , Nam Sung Kim

In the Fully Sharded Data Parallel (FSDP) training pipeline, collective operations can be interleaved to maximize the communication/computation overlap. In this scenario, outstanding operations such as Allgather and Reduce-Scatter can…

Distributed, Parallel, and Cluster Computing · Computer Science 2024-11-12 Mikhail Khalilov , Salvatore Di Girolamo , Marcin Chrapek , Rami Nudelman , Gil Bloch , Torsten Hoefler

Hybrid switching - in which a high bandwidth circuit switch (optical or wireless) is used in conjunction with a low bandwidth packet switch - is a promising alternative to interconnect servers in today's large scale data-centers. Circuit…

Networking and Internet Architecture · Computer Science 2015-12-24 Shaileshh Bojja Venkatakrishnan , Mohammad Alizadeh , Pramod Viswanath
‹ Prev 1 2 3 10 Next ›