Alexandra Fedorova

Gleb Naumenko, Gregory Maxwell, Pieter Wuille et al.

Bitcoin is a top-ranked cryptocurrency that has experienced huge growth and survived numerous attacks. The protocols making up Bitcoin must therefore accommodate the growth of the network and ensure security. Security of the Bitcoin network depends on connectivity between the nodes. Higher connectivity yields better security. In this paper we make two observations: (1) current connectivity in the Bitcoin network is too low for optimal security; (2) at the same time, increasing connectivity will substantially increase the bandwidth used by the transaction dissemination protocol, making it prohibitively expensive to operate a Bitcoin node. Half of the total bandwidth needed to operate a Bitcoin node is currently used to just announce transactions. Unlike block relay, transaction dissemination has received little attention in prior work. We propose a new transaction dissemination protocol, Erlay, that not only reduces the bandwidth consumption by 40% assuming current connectivity, but also keeps the bandwidth use almost constant as the connectivity increases. In contrast, the existing protocol increases the bandwidth consumption linearly with the number of connections. By allowing more connections at a small cost, Erlay improves the security of the Bitcoin network. And, as we demonstrate, Erlay also hardens the network against attacks that attempt to learn the origin node of a transaction. Erlay is currently being investigated by the Bitcoin community for future use with the Bitcoin protocol.

Anand Jayarajan, Jinliang Wei, Garth Gibson et al.

Data parallel training is widely used for scaling distributed deep neural network (DNN) training. However, the performance benefits are often limited by the communication-heavy parameter synchronization step. In this paper, we take advantage of the domain specific knowledge of DNN training and overlap parameter synchronization with computation in order to improve the training performance. We make two key observations: (1) the optimal data representation granularity for the communication may differ from that used by the underlying DNN model implementation and (2) different parameters can afford different synchronization delays. Based on these observations, we propose a new synchronization mechanism called Priority-based Parameter Propagation (P3). P3 synchronizes parameters at a finer granularity and schedules data transmission in such a way that the training process incurs minimal communication delay. We show that P3 can improve the training throughput of ResNet-50, Sockeye and VGG-19 by as much as 25%, 38% and 66% respectively on clusters with realistic network bandwidth