Related papers: Canary: Congestion-Aware In-Network Allreduce Usin…
Ring-based collective operations are widely used in distributed AI training due to their efficient bandwidth utilization. While ring communication excels at pipelining, its performance is heavily dependent on having synchronized step-wise…
The centralized architecture in software-defined network (SDN) provides a global view of the underlying network, paving the way for enormous research in the area of SDN traffic engineering (SDN TE). This research focuses on the load…
As communication protocols evolve, datacenter network utilization increases. As a result, congestion is more frequent, causing higher latency and packet loss. Combined with the increasing complexity of workloads, manual design of congestion…
The evolution of wireless networks and radio access technologies (RATs) has transformed communication from user-driven traffic into a dynamic ecosystem of autonomous systems, including IoT devices, edge nodes, autonomous vehicles, AR/XR…
In this work, we provide the design and implementation of a switch-assisted congestion control algorithm for data center networks (DCNs). In particular, we provide a prototype of the switch-driven congestion control algorithm and deploy it…
Emerging artificial intelligence (AI) and machine learning (ML) workloads present new challenges of managing the collective communication used in distributed training across hundreds or even thousands of GPUs. This paper presents STrack, a…
The method of choice for parameter aggregation in Deep Neural Network (DNN) training, a network-intensive task, is shifting from the Parameter Server model to decentralized aggregation schemes (AllReduce) inspired by theoretical guarantees…
We present a unified analytical framework within which power control, rate allocation, routing, and congestion control for wireless networks can be optimized in a coherent and integrated manner. We consider a multi-commodity flow model with…
Edge machine learning can deliver low-latency and private artificial intelligent (AI) services for mobile devices by leveraging computation and storage resources at the network edge. This paper presents an energy-efficient edge processing…
We focus on the commonly used synchronous Gradient Descent paradigm for large-scale distributed learning, for which there has been a growing interest to develop efficient and robust gradient aggregation strategies that overcome two key…
The interconnection network is a key element in High-Performance Computing (HPC) and Datacenter (DC) systems whose performance depends on several design parameters, such as the topology, the switch architecture, and the routing algorithm.…
This paper presents an asynchronous distributed algorithm to manage multiple trees for peer-to-peer streaming in a flow level model. It is assumed that videos are cut into substreams, with or without source coding, to be distributed to all…
For highly distributed environments such as edge computing, collaborative learning approaches eschew the dependence on a global, shared model, in favor of models tailored for each location. Creating tailored models for individual learning…
In recent years, the issue of energy consumption in high performance computing (HPC) systems has attracted a great deal of attention. In response to this, many energy-aware algorithms have been developed in different layers of HPC systems,…
We consider a network of smart sensors for an edge computing application that sample a time-varying signal and send updates to a base station for remote global monitoring. Sensors are equipped with sensing and compute, and can either send…
Wirelessly connected vehicles that exchange information about traffic conditions can reduce delays caused by congestion. At a 2-to-1 lane reduction, the improvement in flow past a bottleneck due to traffic with a random mixture of 40%…
Being able to identify service slowdowns is crucial to many operational problems. We study how to use observational congestion data to learn service slowdown in a multi-server system that uses adaptive congestion control mechanisms. We show…
Large inter-GPU all-reduce operations, prevalent throughout deep learning, are bottlenecked by communication costs. Emerging heterogeneous architectures are comprised of complex nodes, often containing $4$ GPUs and dozens to hundreds of CPU…
Emerging edge computing paradigms enable heterogeneous devices to collaborate on complex computation applications. However, for congestible links and computing units, delay-optimal forwarding and offloading for service chain tasks (e.g.,…
Communication scheduling has been shown to be effective in accelerating distributed training, which enables all-reduce communications to be overlapped with backpropagation computations. This has been commonly adopted in popular distributed…