Penglai Cui | Searcharxiv

arXiv2022

Enabling Fast and Flexible Distributed Deep Learning with Programmable Switches

Deep learning has been used in a wide range of areas and made a huge breakthrough. With the ever-increasing model size and train-ing data volume, distributed deep learning emerges which utilizes a cluster to train a model in parallel. Unfortunately, the performance is often far from linear speedup due to the communication overhead between cluster nodes. To address this challenge, this paper designs and implements Libra, a network aggregator, that utilizes in-network computation to optimize the communication for distributed DL training in two aspects: 1) reduce active connections and 2) aggregate exchanged network packets. We implemented our Libra on Intel Tofino switches, customized a lightweight host stack and integrated it into an open-source training framework PS-lite. The experimental result shows that our Libra can achieve 1.5~4 times speedup.

cs.NI↗

arXiv2021

NetFC: enabling accurate floating-point arithmetic on programmable switches

In-network computation has been widely used to accelerate data-intensive distributed applications. Some computational tasks, traditional performed on servers, are offloaded to the network (i.e. programmable switches). However, the computational capacity of programmable switches is limited to simple integer arithmetic operations while many of applications require on-the-fly floating-point operations. To address this issue, prior approaches either adopt a float-to-integer method or directly offload computational tasks to the local CPUs of switches, incurring accuracy loss and delayed processing. To this end, we propose NetFC, a table-lookup method to achieve on-the-fly in-network floating-point arithmetic operations nearly without accuracy loss. NetFC adopts a divide-and-conquer mechanism that converts the original huge table into several much small tables together with some integer operations. NetFC adopts a scaling-factor mechanism for computational accuracy improvement, and a prefix-based lossless table compression method to reduce the memory overhead. We use different types of datasets to evaluate NetFC. The experimental results show that the average accuracy of NetFC can be as high as up to 99.94% at worst with only 448KB memory consumption. Furthermore, we integrate NetFC into Sonata for detecting Slowloris attack, yielding significant decrease of detection delay.

cs.NI↗