Internet traffic bursts usually happen within a second, thus conventional burst mitigation methods ignore the potential of Traffic Engineering (TE). However, our experiments indicate that a TE system, with a sub-second control loop latency, can effectively alleviate burst-induced congestion. TE-based methods can leverage network-wide tunnel-level information to make globally informed decisions (e.g., balancing traffic bursts among multiple paths). Our insight in reducing control loop latency is to let each router make local TE decisions, but this introduces the key challenge of minimizing performance loss compared to centralized TE systems. In this paper, we present RedTE, a novel distributed TE system with a control loop latency of < 100ms, while achieving performance comparable to centralized TE systems. RedTE's innovation is the modeling of TE as a distributed cooperative multi-agent problem, and we design a novel multi-agent deep reinforcement learning algorithm to solve it, which enables each agent to make globally informed decisions solely based on local information. We implement real RedTE routers and deploy them on a WAN spanning six city datacenters. Evaluation reveals notable improvements compared to existing solutions: < 100ms of control loop latency, a 37.4% reduction in maximum link utilization, and a 78.9% reduction in average queue length.
28 November 2023 in IEEE Transactions on Services Computing
Federated Learning (FL) has emerged as a powerful paradigm in Artificial Intelligence, facilitating the parallel training of Artificial Neural Networks on edge devices while safeguarding data privacy. Nonetheless, to encourage widespread adoption, Federated Learning Frameworks (FLFs) must tackle (i) the power imbalance between a central authority and its participants, and (ii) the challenge of equitably measuring and incentivizing contributions. Existing approaches to decentralize and incentivize FL processes are hindered by (i) computational overhead and (ii) uncertainty in contribution assessment [1]), limiting FL's scalability beyond use cases where trust between participants and the server is established. This work introduces a cutting-edge, blockchain-enabled federated learning framework that incorporates Federated Knowledge Distillation (FD) with compressed 1-bit soft-labels, aggregated through a smart contract. Furthermore, we present the Peer Truth Serum for Federated Distillation (PTSFD), which cultivates an incentive-compatible ecosystem by rewarding honest participation based on an implicit yet effective comparison of worker contributions. The primary innovation stems from its lightweight architecture that simultaneously promotes decentralization and incentivization, addressing critical challenges in contemporary FL approaches.
19 June 2020 in IEEE/ACM Transactions on Networking
In large-scale distributed machine learning (DML), the network performance between machines significantly impacts the speed of iterative training. In this paper we propose BML , a scalable, high-performance and fault-tolerant DML network architecture on top of Ethernet and commodity devices. BML builds on BCube topology, and runs a fully-distributed gradient synchronization algorithm. Compared to a Fat-Tree network with the same size, a BML network is expected to take much less time for gradient synchronization, for both low theoretical synchronization time and its benefit to RDMA transport. With server/link failures, the performance of BML degrades in a graceful way. Experiments of MNIST and VGG-19 benchmarks on a testbed with 9 dual-GPU servers show that, BML reduces the job completion time of DML training by up to 56.4% compared with Fat-Tree running state-of-the-art gradient synchronization algorithm.
A numerical simulation is conducted to investigate effects of dimple locations on the flow structure and heat transfer in a dimpled-pin finned channel. Pin fins are arranged with a staggered layout on the endwall surface. The dimple print diameter is the same as the pin fin diameter. Five different dimple locations are investigated in this study. All results are acquired at Reynolds numbers ranging from 8,500 to 60,000. The results show that the dimple locations have remarkable effects on the flow structure and heat transfer in a dimpled-pin finned channel. The Nusselt number of the dimpled-pin finned channel is only increased by 3.52%, while the friction factor is increased by 1.81% compared to the pure pin finned channel when the dimple is located in the wake region. The wake and reversed flow are responsible for this low heat transfer augmentation. When the dimple is located in between the two pin fins, the counter-rotating vortex and impingement within the dimple produce the highest Nusselt number and the lowest friction factor among the tested dimple locations. In this case, the Nusselt number is increased by 20.94% and the friction factor is decreased by 1.69% compared to the pure pin finned channel.
21 November 2017 in Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science
In an unshrouded high-pressure turbine, the upstream vane wake, vane–blade interaction and blade tip leakage flow indicate complex and unsteady flow characteristics. Considering a high-pressure turbine stage of low-aspect ratio, the effects of the flat tip, cavity tip and winglet-cavity tip on the unsteady flow characteristics are investigated by numerical simulation. The exit Mach number and Reynolds number based on the chord of vane are 0.9 and 5.5×105, respectively. The pressure ratio of stage is 2.4. The time-resolved results indicate that the winglet-cavity tip scheme has smaller time variation of the total performance parameters and obtains a smaller tip leakage mass flow and a higher turbine stage efficient than the other cases. The vane–rotor interaction affects the pressure distribution at the region of blade leading edge remarkably and leads to a large time variation of tip leakage mass flow at the front of blade tip. The unsteady aerodynamic performance is analyzed with entropy-increase distribution at stage outlet and the vane–rotor interaction is discussed by the entropy-increase phase–time and phase–phase diagrams at blade outlet. Compared with the flat tip and cavity tip, the upper passage vortex loss is reduced obviously by the winglet-cavity tip. Thus, it is evaluated that the improvement of turbine stage coefficient with winglet-cavity tip results from the reduction of the upper passage vortex loss.
The differences of flow field in bowed blade cascade and that in straight blade cascade are systematically studied in this paper. To bow a blade means to change its geometric boundary condition. This change not only affect the pressure distribution along the blade profile exit Mach number but also has great effect on the original position and development of the passage vertex. All of the changes mentioned above have great influence on the loss. Numerical simulation result showed that blade bowing can decrease the cross-pressure gradient near the end wall. This trend will be more obvious with the increase of the bow angle. The pressure gradient decrease is beneficial to weaken the passage vortex strength and reduces the secondary loss near the endwalls. In addition, Pressure gradient from endwalls to midspan can be established near suction surface in positively bowed blade. With the increase of bow angle, this C-type pressure distribution is remarkable. It is also found that this C-type pressure distribution will influence the position of corner vortex near the suction surface and will also influence the position and size of the passage vortex. Blade bowing also has great influence on the position of the saddle point near the leading edge and the separated line of the horseshoe vortex. It is found that the position of the saddle point and the separated line of both legs of the horseshoe vortex move forward in a positively bowed blade. The passage vortex structure in bowed cascade is also presented. It can be concluded that a bowed blade can make the passage vortex stable and helps change its structure from loose to compact. Blade bowing is also beneficial to limit the influence domain of the unstable passage vortex core by the stable limit cycle.