Parallel Processing Letters

The size of deep neural networks (DNNs) grows rapidly as the complexity of the machine learning algorithm increases. Distributed deep learning based on model parallelism has been widely used to satisfy the requirements of DNN training related to computation and memory. In this paper, we propose a training framework for pipeline parallelism called BaPipe (Balanced Pipeline) that can automatically explore methods to schedule pipeline parallelism and balanced partition strategies for DNN training on heterogeneous accelerator clusters. In BaPipe, each accelerator calculates the forward and backward propagation for the assigned partition of networks to implement an intra-batch pipeline parallelism strategy. By considering the parameters of DNN models as well as the computation, memory, and communication resources of each accelerator, BaPipe automatically selects the most suitable method of pipeline scheduling from among multiple proposed scheduling modes. It also uses a novel strategy to automatically investigate load balancing in the context of inter-layer partition, intra-layer partition, and coarse-grained partition. We trained such DNNs as VGG-16, ResNet-50, and Google’s Neural Machine Translation (GNMT) on GPU clusters, and simulated the training-related performance of FPGA clusters. Compared with the state-of-the-art frameworks for data parallelism (DP) and pipeline parallelism, BaPipe provides a speedup of $3.2\times$ and $4\times$ of memory reduction on various homogeneous and heterogeneous platforms.

Wireless Sensor Networks (WSN) consists of numerous of low cost and less-energy sensor nodes that are responsible to gather and transmit the data packets from one node to destination point. WSN has a wide range of applications over agriculture, military, traffic monitoring, instrument surveillance, and security monitoring. In WSN, the nodes are located in a specific region to create a wireless network. The effective data communication among sensors is a challenging task because of different complex parameters. Typically, clustering is a well-preferred methodology to provide the effective communication by partitioning the nodes into different clusters. Every cluster possesses individual cluster head that transmits the data to other sensor nodes. Therefore, it is substantial to choose optimal cluster head and optimal route for effective transmission with less energy consumption and less delay. To increase the network efficiency and sink utilization, an energy aware routing algorithm called Fractional Competitive Fruit Fly Optimizer (FrCFFO) is designed, which is an integration of Fractional concept into the Competitive Fruit Fly Optimizer (CFFO). Here, the energy prediction is performed using Deep Quantum Neural Network (QNN). Effective CH selection and routing is done using the proposed FrCFFO and the fitness parameter is considered depending upon the factors like energy, distance, link lifetime, trust, and delay. Moreover, the developed FrCFFO has achieved effective performance with minimum delay of 0.098sec, maximum energy of 0.233J, and maximum PDR of 90.81%.

Our manuscript analyzes the performance of supply chain of marginal cost-intensive green products under random demand. This model contains two members: one manufacturer and one retailer. We study how different parameters influence performance of both centralized and decentralized channel. Our results suggest that: (1) overall profit in centralized channel is always higher comparing to decentralized channel; (2) centralized channel has higher production quantity and lower retail price; (3) the green-level is equal in these two channels; (4) in decentralized channel, the retailer will get more profits than the manufacturer.

The cyclic vertex (resp., edge) connectivity of a graph $G$, denoted by ${\kappa }_c(G)$ (resp., ${\lambda }_c(G)$), is the minimum number of vertices (resp., edges) whose removal from $G$ results in a disconnected graph and at least two remaining components contain cycles. Thus, to determine the exact values of ${\kappa }_c(G)$ and ${\lambda }_c(G)$ is important in the reliability assessment of interconnection networks. However, the study of the cyclic vertex (edge) connectivity is less involved. In this paper, we determine the cyclic vertex (edge) connectivity of the burnt pancake graphs $B{P}_n$ which is the Cayley graph of the group of signed permutations using prefix reversals as generators. By exploring the combinatorial properties and fault-tolerance of $B{P}_n$, we show ${\kappa }_c(B{P}_n)=8(n-2)$ and ${\lambda }_c(B{P}_n)=8(n-2)$ for $n\ge 3$. Moreover, we determine that ${\lambda }^2(B{P}_n)=8(n-2)$ for $n\ge 3$.

A mobile ad hoc network (MANET) is a set of autonomous mobile devices connected by wireless links in a distributed manner and without a fixed infrastructure. Mobile Ad-Hoc Networks (MANETs) is considered as an infrastructure-less and constant self-configured network of wireless devices. Here, multicast has emerged as an efficient tool for communication. Recently, distinct research works have been studied in wireless MANETs for designing multicast routing protocols. Multicasting is a procedure used for allowing the users to send the same messages simultaneously to a set of users. Though, it suffers from different problems regarding implementation in ad-hoc networks owing to its shorter battery lifetime, and lack of bandwidth dynamic nature of the mobile devices. Hence, the main aim of this paper is to propose a new QoS-aware multi-cast routing protocol in MANET using the hybrid meta-heuristic algorithm. The proposed routing protocol will focus on the development of a hybrid meta-heuristic-based “reliable neighbor nodes selection” scheme over the Multicast Ad-Hoc on Demand Distance Vector routing protocol (MAODV) protocol. The “reliable neighbor nodes selection” is accomplished by Fitness sorted Red Deer-Cat Swarm Optimization (FRD-CSO) algorithm focusing on the constraints like energy model, mobility model, and reliability pair factor. Once the optimal node selection is performed, the proposed QoS-aware protocol is validated by the route discovery mechanism and route reply process. The simulation has been performed to show the efficiency of the designed multicast routing over other multicast routing protocols that demonstrates the noteworthy enhancement in terms of diverse performance measures.

The prefix sums problem for an array $X=(x_0,x_1,\dots ,x_{n-1})$ is to compute all sums $x_0+x_1\cdots +x_i$, $\forall 0\le i<n$. In this paper, we introduce an improvement for the best previous algorithm that runs in $O({log}^{\ast }n)$ time using $O(n/{log}^{\ast }n)$ processors on a Sum Concurrent Read Concurrent Write, Parallel Random Access Machine (Sum-CRCW PRAM). The improvements include (1) reducing the total number of operations and (2) reducing the amount of storage required by the algorithm.