High performance communications in processor networks

In order to provide an arbitrary and fully dynamic connectivity in a network of processors, transport mechanisms must be implemented, which provide the propagation of data from processor to processor, based on addresses contained within a packet of data. Such data transport mechanisms must satisfy a number of requirements, namely deadlock and livelock freedom, good hot-spot performance, high throughput and low latency. The paper proposes a solution to these problems, which allows deadlock free, adaptive, high throughput packet routing to be implemented on networks of processors. Examples are given which illustrate the technique for two-dimensional (2D) array and toroidal networks. The implementation of this scheme on arrays of transputers is considered. The scheme also serves as a basis for very-low latency routing strategy introduced here as well.

Training set parallelism and network based parallelism are two popular paradigms for parallelizing a feedforward (artificial) neural network. Training set parallelism is particularly suited to feedforward neural networks with backpropagation learning where the size of the training set is large in relation to the size of the network. This paper analyzes training set parallelism for feedforward neural networks when implemented on a transputer array configured in a pipelined ring topology. Theoretical expressions for the time per epoch (iteration) and optimal size of a processor network are derived when the training set is equally distributed among the processing nodes. These show that the speed up is a function of the number of patterns per processor, communication overhead per epoch and the total number of processors in the topology. Further analysis of how to optimally distribute the training set on a given processor network when the number of patterns in the training set is not an integer multiple of the number of processors, is also carried out. It is shown that optimal allocation of patterns in such cases is a mixed integer programming problem. Using this analysis it is found that equal distribution of training patterns among the processors is not the optimal way to allocate the patterns even when the training set is an integer multiple of the number of processors. Extension of the analysis to processor networks comprising processors of different speeds is also carried out. Experimental results from a T805 transputer array are presented to verify all the theoretical results.

In recent times the work on networks of processors has become very important, due to the low cost and the availability of these systems. This is why it is interesting to study algorithms on networks of processors. In this paper we study on networks of processors different Eigenvalue Solvers. In particular, the Power method, deflation, Givens algorithm, Davidson methods and Jacobi methods are analized using PVM and MPI. The conclusion is that the solution of Eigenvalue Problems can be accelerated by using networks of processors and typical parallel algorithms, but the high cost of communications in these systems gives rise to small modifications in the algorithms to achieve good performance.

The 5G user plane function (UPF) is a critical inter-connection point between the data network and cellular network infrastructure. It governs the packet processing performance of the 5G core network. UPFs also need to be flexible to support several key control plane operations. Existing UPFs typically run on general-purpose CPUs, but have limited performance because of the overheads of host-based forwarding. We design Synergy, a novel 5G UPF running on SmartNICs that provides high throughput and low latency. It also supports monitoring functionality to gather critical data on user sessions for the prediction and optimization of handovers during user mobility. The SmartNIC UPF efficiently buffers data packets during handover and paging events by using a two-level flow-state access mechanism. This enables maintaining flow-state for a very large number of flows, thus providing very low latency for control and data planes and high throughput packet forwarding. Mobility prediction can reduce the handover delay by pre-populating state in the UPF and other core NFs. Synergy performs handover predictions based on an existing recurrent neural network model. Synergy's mobility predictor helps us achieve 2.32× lower average handover latency. Buffering in the SmartNIC, rather than the host, during paging and handover events reduces packet loss rate by at least 2.04×. Compared to previous approaches to building programmable switch-based UPFs, Synergy speeds up control plane operations such as handovers because of the low P4-programming latency leveraging tight coupling between SmartNIC and host.

Communication in industrial wireless networks necessitates reliability and precision. Besides, the existence of interference or traffic in the network must not affect the estimated network properties. Therefore, data packets have to be sent within a certain time frame and over a reliable connection. However, the working scenarios and the characteristics of the network itself make it vulnerable to node or link faults, which impact the transmission reliability and overall performance. This article aims to introduce a developed multipath routing model, which leads to cost-effective planning, low latency and high reliability of industrial wireless mesh networks, such as the WirelessHART networks. The multipath routing model has three primary paths, and each path has a backup node. The backup node stores the data transmitted by the parent node to grant communication continuity when primary nodes fail. The multipath routing model is developed based on optimal network planning and deployment algorithm. Simulations were conducted on a WirelessHART simulator using Network Simulator (NS2). The performance of the developed model is compared with the state-of-the-art. The obtained results reveal a significant reduction in the average network latency, low power consumption, better improvement in expected network lifetime, and enhanced packet delivery ratio which improve network reliability.

Intelligent transportation systems (ITS) in smart cities offer extensive network coverage and simultaneous connections, even in high-mobility scenarios within the IoT network. This way, 5G technology will become a key enabler for IoT and ITS by allowing massive vehicular connections, high throughput, low latency, and wide coverage. Thus, resource sharing becomes very important in dealing with massive IoT connections in ITS and providing efficient communication among vehicles for smart cities. We propose a resource-sharing scheme for Cellular-vehicle-to-everything (C-V2X) communication in 5G cellular networks for smart cities, having two groups of users: Vehicle-to-Vehicle (V2V) and Vehicle-to-Infrastructure (V2I). Our primary goal is to maximise the throughput of V2V pairs and V2I users while preserving the QoS for both users. The maximisation problem formulated is non-convex and is solved using the Lagrange multiplier method by optimizing the transmit power of both V2V and V2I users. The overall system performance is evaluated in terms of throughput of V2V and V2I users for different SINR, number of vehicles, velocities, and transmit power of vehicular users. Furthermore, the EE of V2V communication is analyzed in terms of the transmit power of VUs. Also, the proposed resource-sharing scheme is compared with the existing schemes, demonstrated through simulations.

Drawing inspiration from several previous projects, we present an ownership-record-free software transactional memory (STM) system that combines extremely low overhead with unusually clean semantics. While unlikely to scale to hundreds of active threads, this system offers many appealing features: very low fast-path latency--as low as any system we know of that admits concurrent updates; publication and privatization safety; livelock freedom; a small, constant amount of global metadata, and full compatibility with existing data structure layouts; no false conflicts due to hash collisions; compatibility with both managed and unmanaged languages, and both static and dynamic compilation; and easy acccommodation of closed nesting, inevitable (irrevocable) transactions, and starvation avoidance mechanisms. To the best of our knowledge, no extant STM system combines this set of features.While transactional memory for processors with hundreds of cores is likely to require hardware support, software implementations will be required for backward compatibility with current and near-future processors with 2--64 cores, as well as for fall-back in future machines when hardware resources are exhausted. Our experience suggests that NOrec may be an ideal candidate for such a software system. We also observe that it has considerable appeal for use within the operating system, and in systems that require both closed nesting and publication safety.

Drawing inspiration from several previous projects, we present an ownership-record-free software transactional memory (STM) system that combines extremely low overhead with unusually clean semantics. While unlikely to scale to hundreds of active threads, this "NOrec" system offers many appealing features: very low fast-path latency--as low as any system we know of that admits concurrent updates; publication and privatization safety; livelock freedom; a small, constant amount of global metadata, and full compatibility with existing data structure layouts; no false conflicts due to hash collisions; compatibility with both managed and unmanaged languages, and both static and dynamic compilation; and easy acccommodation of closed nesting, inevitable (irrevocable) transactions, and starvation avoidance mechanisms. To the best of our knowledge, no extant STM system combines this set of features. While transactional memory for processors with hundreds of cores is likely to require hardware support, software implementations will be required for backward compatibility with current and near-future processors with 2--64 cores, as well as for fall-back in future machines when hardware resources are exhausted. Our experience suggests that NOrec may be an ideal candidate for such a software system. We also observe that it has considerable appeal for use within the operating system, and in systems that require both closed nesting and publication safety.

Seamless handoff support is an essential issue to ensure continuous communications in wireless mesh networks (WMNs). Due to the existence of multi-hop wireless links, traditional handoff schemes designed for single-hop wireless access networks can hardly guarantee the low handoff latency requirement in WMNs. Existing solutions on reducing the handoff delay in WMNs ignore one important factor for the long handoff delay: the channel access delay of handoff signaling packets over the multi-hop wireless mesh backbone network. Various WMN channel assignment algorithms are designed to optimize the performance of data packets, but they do not address how to resolve the contentions between data packets and handoff signaling packets. In this paper, we propose a novel channel assignment algorithm in WMNs to reduce the channel access delay of handoff signaling packets over multihop wireless links. In addition, handoff solutions under the new channel assignment scheme are proposed. OPNET simulation results show that handoff performance can be improved by our proposed handoff schemes, as compared to existing solutions.

Blockchain is a new technology for processing complex and disordered information with respect to business and other industrial applications. This work is aimed at studying the consensus algorithm of blockchain to improve the performance of blockchain. Despite their advantages, the proof of stake (POS) algorithm and the practical Byzantine fault tolerance (PBFT) algorithm have high latency, low throughput, and poor scalability. In this paper, a blockchain hybrid consensus algorithm which combines advantages of the POS and PBFT algorithms is proposed, and the algorithm is divided into two stages: sortition and witness. The proposed algorithm reduces the number of consensus nodes to a constant value by verifiable pseudorandom sortition and performs transaction witness between nodes. The algorithm is improved and optimized from three dimensions: throughput, latency, and scalability. The experimental results show that the improved hybrid consensus algorithm is significantly superior to the previous single algorithms for its excellent scalability, high throughput, and low latency.

There is an important class of applications in data centre called online dataintensive (OLDI) application, such as search engine, electronic shopping, and advertising, which has low latency, high throughput, and high burstiness. For better user experience, the query response delay in the data centre is controlled under 300 milliseconds. DCTCP can better meet the requirements of OLDI, but all the switches in the data centre are required to support RED. For data centre that does not fully support RED switch, TCP protocol can only be used. TCP Incast will appear in the application, which is far from the requirements of OLDI application. This paper proposes the TCP protocol of data centre based on window control. It is in existing switches and network devices,and centralized control makes the send window match with the switch queue cache. In the case of ensuring low latency, it keeps the switch queue in a small range of fluctuation, enabling the data centre to meet low latency, high throughput and high burstiness application requirements. Simulation experiments verify that WCTCP congestion control method can effectively solve the above problems. It is 5 times lower than TCPNewReno packet loss rate, 6 times higher than the network utilization rate, and the network performance is close to the DCTCP protocol.

High throughput and low latency data processing is essential for systems requiring live decision making, control, and machine learning-optimized data reduction. We focus on two distinct use cases for in-flight streaming data processing for a) X-ray pulse reconstruction at SLAC's LCLS-II Free-Electron Laser and b) control diagnostics at the DIII-D tokamak fusion reactor. Both cases exemplify high throughput and low latency control feedback and motivate our focus on machine learning at the edge where data processing and machine learning algorithms can be implemented in field programmable gate array based hardware immediately after the diagnostic sensors. We present our recent work on a data preprocessing chain which requires fast featurization for information encoding. We discuss several options for such algorithms with the primary focus on our discrete cosine and sine transform-based approach adapted for streaming data. These algorithms are primarily aimed at implementation in field programmable gate arrays, favoring linear algebra operations that are also aligned with the recent advances in inference accelerators for the computational edge.

In this paper, we investigate the throughput performance of single-packet and multi-packet hybrid-automatic repeat request (HARQ) with blanking for downlink non-orthogonal multiple access (NOMA) systems. While conventional single-packet HARQ achieves high throughput at the expense of high latency, multi-packet HARQ, where several data packets are sent in the same channel block, can achieve high throughput with low latency. Previous works have shown that multi-packet HARQ outperforms single-packet HARQ in orthogonal multiple access (OMA) systems, especially in the moderate to high signal-to-noise ratio regime. This work amalgamates multi-packet HARQ with NOMA to achieve higher throughput than the conventional single-packet HARQ and OMA, which has been adopted in the legacy mobile networks. We conduct theoretical analysis for the throughput per user and also investigate the optimization of the power and rate allocations of the packets, in order to maximize the weighted-sum throughput. It is demonstrated that the gain of multi-packet HARQ over the single-packet HARQ in NOMA systems is reduced compared to that obtained in OMA systems due to inter-user interference. It is also shown that NOMA-HARQ cannot achieve any throughput gain with respect to OMA-HARQ when the error propagation rate of the NOMA detector is above a certain threshold.

In this paper, the optimal solution of differential algebraic equation (DAE) systems over a network of processors is investigated. A sliding-mode control (SMC) approach is used to enforce the algebraic constraints of the DAEs. Time delays associated with network communication give rise to a sliding-control problem with sensor time delays. A new approach is presented that decouples this problem into two independent problems. Using this property, a mixed H <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">infin</sub> optimization approach is applied to minimize errors. The results are applied to solve the problem of the optimal bandwidth allocation between the processors under the constraint of limited communication bandwidth. Experimental results for a distributed simulation application on a TDMA network demonstrate the validity of the approach

Motivated by the emerging standards for indoor pico-cellular wireless systems, such as the Bluetooth, we propose and study the scheduling policies for master driven time division duplex (TDD) wireless networks. In these networks, the frequency band is divided into time slots, and each end (i.e., master or slave) takes turns in using the time slots. In Bluetooth, a slave transmits packets in the reverse slot only after the master polls the slave in a forward slot (by sending data to it). The conventional scheduling policies such as round robin do not perform well in these systems as they are not suited to the tight coupling of the uplink-downlink. We propose new scheduling policies, (i) the priority scheme, and, (ii) the K-fairness scheme that utilize the state at the master and slaves to schedule the TDD slots effectively. Active slaves are differentiated based upon the binary information (i.e., the presence or absence of packets in a slave queue) about the master-slave queue pairs. The priority scheme achieves high throughput as compared to the packet-by-packet generalized processor sharing (PGPS) based policies while guaranteeing a minimal service to each active slave while the K-fairness policy is characterized by a tight fairness bound as well as high system throughput. We then extend these policies for scheduling variable size data in the presence of voice. Further, since Bluetooth supports variable size data packets (1, 3 or 5 slots) on the same connections, the segmentation and reassembly (SAR) can significantly impact scheduling of data packets by varying packet size distribution. We propose an intelligent SAR policy (ISAR) and compare it with the naive random-SAR in which the data packet sizes (i.e., 1, 3 or 5) are assigned probabilistically. ISAR adapts MAC packet size at the master and slave queues depending on the data arrival rates at both the queues.