Related papers: The Swapped Dragonfly
The Swapped Dragonfly with M routers per group and K global ports per router is denoted D3(K;M) [1]. It has n=KMM routers and is a partially populated Dragonfly. A Swapped Dragonfly with K and M restricted is studied in this paper. There…
Existing high-performance computing (HPC) interconnection architectures are based on high-radix switches, which limits the injection/local performance and introduces latency/energy/cost overhead. The new wafer-scale packaging and high-speed…
DPillar has recently been proposed as a server-centric datacenter network and is combinatorially related to (but distinct from) the well-known wrapped butterfly network. We explain the relationship between DPillar and the wrapped butterfly…
The Dragonfly topology is currently one of the most popular network topologies in high-performance parallel systems. The interconnection networks of many of these systems are built from components based on the InfiniBand specification.…
Much like classical supercomputers, scaling up quantum computers requires an optical interconnect. However, signal attenuation leads to irreversible qubit loss, making quantum interconnect design guidelines and metrics different from…
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,…
High-radix interconnects such as Dragonfly and its variants rely on adaptive routing to balance network traffic for optimum performance. Ideally, adaptive routing attempts to forward packets between minimal and non-minimal paths with the…
We consider single-source single-sink (ss-ss) multi-hop networks, with slow-fading links and single-antenna half-duplex relays. We identify two families of networks that are multi-hop generalizations of the well-studied two-hop network:…
The explosively growing communication traffic in datacenters imposes increasingly stringent performance requirements on the underlying networks. Over the last years, researchers have developed innovative optical switching technologies that…
Convolutional neural networks have become the main tools for processing two-dimensional data. They work well for images, yet convolutions have a limited receptive field that prevents its applications to more complex 2D tasks. We propose a…
We introduce a new kind of linear transform named Deformable Butterfly (DeBut) that generalizes the conventional butterfly matrices and can be adapted to various input-output dimensions. It inherits the fine-to-coarse-grained learnable…
Most deep neural networks (DNNs) consist fundamentally of convolutional and/or fully connected layers, wherein the linear transform can be cast as the product between a filter matrix and a data matrix obtained by arranging feature tensors…
With the idea of an eventual classification of 3-bridge links,\ we define a very nice class of 3-balls (called butterflies) with faces identified by pairs, such that the identification space is $S^{3},$ and the image of a prefered set of…
We study in this paper a two-hop relaying network consisting of one source, one destination, and three amplify-and-forward (AF) relays operating in a half-duplex mode. In order to compensate for the inherent loss of capacity pre-log factor…
Multi-dimensional discrete Fourier transforms (DFT) are typically decomposed into multiple 1D transforms. Hence, parallel implementations of any multi-dimensional DFT focus on parallelizing within or across the 1D DFT. Existing DFT packages…
Several interconnection networks are based on the complete graph topology. Networks with a moderate size can be based on a single complete graph. However, large-scale networks such as Dragonfly and HyperX use, respectively, a hierarchical…
Suppose we concatenate two directed graphs, each isomorphic to a $d$ dimensional butterfly (but not necessarily identical to each other). Select any set of $2^k$ input and $2^k$ output nodes on the resulting graph. Then there exist node…
Software Defined Networks has seen tremendous growth and deployment in different types of networks. Compared to traditional networks it decouples the control logic from network layer devices, and centralizes it for efficient traffic…
Executing deep neural networks (DNNs) on edge artificial intelligence (AI) devices enables various autonomous mobile computing applications. However, the memory budget of edge AI devices restricts the number and complexity of DNNs allowed…
Software-defined networking (SDN) attracts the attention of the research community in recent years, as evidenced by a large number of survey and review papers. The architecture of SDN clearly recognizes three planes: application, control,…