Network and Parallel Computing: 17th IFIP WG 10.3 International Conference, NPC 2020, Zhengzhou, China, September 28–30, 2020, Revised Selected Papers (Theoretical Computer Science and General Issues)

This book constitutes the proceedings of the 17th IFIP WG 10.3 International Conference on Network and Parallel Computing, NPC 2020, held in Zhengzhou, China, in September 2020. The 34 full and 7 short papers presented in this volume were carefully reviewed and selected from 95 submissions. They were organized in topical sections named: accelerator; AI; algorithm; architecture and hardware; big data and cloud; edge computing; emerging; network; and storage. Preface Organization Contents Accelerator Compiler-Assisted Operator Template Library for DNN Accelerators 1 Introduction 2 Design 2.1 User-View Abstraction 2.2 Memory Abstraction 2.3 Computation Abstraction 3 Implementation 3.1 Expression Template 3.2 Compiler-Assisted Optimizations 4 Performance Evaluation 4.1 Benchmarks and Baselines 4.2 Experimental Results 5 Related Work 6 Conclusion References A Dynamic Mapping Model for General CNN Accelerator Based on FPGA 1 Introduction 2 Related Work 3 CNN Basics and Matrix Multiplication 3.1 2D and 3D Convolution 3.2 Dilated Convolution 3.3 Mapping Convolutions to Matrix Multiplications 4 Accelerator Architecture Design 4.1 MAC Array and Buffer Setting 4.2 Dynamic Mapping Model 5 Evaluation 5.1 Experimental Setup 5.2 Experimental Results 6 Conclusions References A Dynamic Protection Mechanism for GPU Memory Overflow 1 Introduction 2 Background 2.1 CUDA Programming Model 2.2 Unified Memory 2.3 GPU Allocator Optimization 3 Design Overview 3.1 Buffer Structure 3.2 Overview of the System 4 Implementation 4.1 Memory Allocation 4.2 Buffer Structure Construction 4.3 Buffer Overflow Detection 5 Evaluation 5.1 Experiment Setup 5.2 Performance Analysis 6 Conclusion References AI MTLAT: A Multi-Task Learning Framework Based on Adversarial Training for Chinese Cybersecurity NER 1 Introduction 2 Related Work 2.1 NER 2.2 Adversarial Training 3 Methodology 3.1 Adversarial Training 3.2 Encoding Layer 3.3 Training 4 Experiments 4.1 Datasets 4.2 Comparison Methods 4.3 Hyper-Parameter Settings 5 Results and Analysis 5.1 Results on the Chinese Cybersecurity NER Dataset 5.2 Results on Chinese General NER Datasets 5.3 Case Study 6 Conclusion References Learning-Based Evaluation of Routing Protocol in Vehicular Network Using WEKA 1 Introduction 2 Related Work 3 Proposed Solution 3.1 Improved Protocol Description 3.2 Machine Learning Algorithms Description 4 Evaluation Results 5 Conclusion References Accelerating Large-Scale Deep Convolutional Neural Networks on Multi-core Vector Accelerators 1 Introduction 2 Background and Related Work 2.1 The Architecture of Vector Accelerator 2.2 Related Work 3 Algorithm Mapping of CNN on Vector Accelerator 3.1 Computation Method for Converting Convolution Operations to Large-Scale Matrix Multiplication 3.2 Vectorization Method of Valid Convolution Layer 3.3 Vectorization Method of Same Convolution Layer 3.4 Vectorization Method of Pooling Layer 3.5 Vectorization Method of Fully Connected Layer 4 Experimental Results and Performance Analysis 4.1 Performance of Convolution Layers 4.2 Performance of Fully Connected Layers 4.3 Performance of Pooling Layers 4.4 Performance Comparison 5 Conclusion References M-DRL: Deep Reinforcement Learning Based Coflow Traffic Scheduler with MLFQ Threshold Adaption 1 Introduction 2 Coflow Scheduling Model Without Prior Knowledge 2.1 Background 2.2 Coflow Scheduling Based on MLFQ 3 Data-Driven MLFQ Threshold Setting Using DRL 3.1 DRL Formulation 3.2 Algorithm Design 4 Evaluation 4.1 Workload 4.2 Evaluation Metrics 4.3 Simulation Result 5 Related Work 6 Conclusion References A Close Look at Multi-tenant Parallel CNN Inference for Autonomous Driving 1 Introduction 2 Background and Related Work 3 Design of MPInfer 3.1 Overview of MPInfer 3.2 Design Goals 3.3 Design Issues 4 Evaluation 4.1 Experimental Settings 4.2 Experimental Results 5 Conclusion References A Multi-model Super-Resolution Training and Reconstruction Framework 1 Introduction 2 Related Work 2.1 Super-Resolution 2.2 Image Classification 3 Multi-model SR Training and Reconstruction Framework 3.1 Pruning Layer 3.2 Classification Layer 3.3 Fusion Layer 4 Experiments 4.1 Environment Setup 4.2 Experiment Details 5 Conclusion References Deep Visible and Thermal Image Fusion with Cross-Modality Feature Selection for Pedestrian Detection 1 Introduction 2 Proposed Method 2.1 Two-Branch Feature Extraction Backbone Network with Squeeze-Excitation Module 2.2 Cross-Modality Feature Selection Module 2.3 Optimization 3 Experimental Results 3.1 Experimental Setup 3.2 Comparison with the State-of-the-Art Methods 3.3 Ablation Study 4 Conclusion References LCache: Machine Learning-Enabled Cache Management in Near-Data Processing-Based Solid-State Disks 1 Introduction 2 Background and Related Work 2.1 Near-Data Processing Storage Device 2.2 Cache Management 3 Overview System 4 System Implementation 4.1 Machine Learning Module 4.2 Proactive Update Function 4.3 Read and Write Operation 5 Experimental Setup and Evaluation 5.1 Experimental Setup 5.2 Performance Evaluation 6 Conclusion References Security Situation Prediction of Network Based on Lstm Neural Network 1 Introduction 2 ONS2P Algorithm 3 Experiment 4 Related Works 5 Conclusion References Algorithm Dynamic GMMU Bypass for Address Translation in Multi-GPU Systems 1 Introduction 2 Background 2.1 Programming Models 2.2 Remote Data Access Mechanisms 2.3 Address Translation Architectures 3 Motivation 4 GMMU Bypass 5 Methodology 5.1 Experimental Setup 5.2 Workloads 6 Evaluation 6.1 Performance 6.2 Analysis 6.3 Hardware Cost 7 Related Work 7.1 Address Translation on GPU 7.2 Performance Optimization for Multi-GPU Systems 8 Conclusion References Parallel Fast DOA Estimation Algorithm Based on SML and Membrane Computing 1 Introduction 2 Mathematical Model and SML 2.1 Array Signal Model 2.2 Stochastic Maximum Likelihood Algorithm 3 SML Based on Membrane Computing 3.1 Membrane Division of the Solution Space 3.2 Intra-membrane Local Search Parallel Algorithm 3.3 Global Optimization Strategy 3.4 DOA Estimation Step Based on Membrane Computing 4 Simulation and Performance Analysis 5 Conclusion References Segmented Merge: A New Primitive for Parallel Sparse Matrix Computations 1 Introduction 2 Related Work 2.1 Parallel Segmented Sum 2.2 Parallel Segmented Scan 2.3 Parallel Segmented Sort 3 Segmented Merge and Its Parallel Algorithm 3.1 Definition of Segmented Merge 3.2 Serial Algorithm for Segmented Merge 3.3 Simple Parallel Algorithm for Segmented Merge 3.4 Improved Parallel Algorithm for Segmented Merge on GPU 4 Performance Evaluation 4.1 Experimental Setup 4.2 Performance of SpTRANS Using Segmented Merge 4.3 Performance of SpGEMM Using Segmented Merge 5 Conclusion References A Hierarchical Model of Control Logic for Simplifying Complex Networks Protocol Design 1 Introduction 2 Motivation and Related Works 2.1 Motivation 2.2 Related Works 3 DoubleDeck Model Overview 4 Experiment 5 Conclusion References Architecture and Hardware FPGA-Based Multi-precision Architecture for Accelerating Large-Scale Floating-Point Matrix Computing 1 Introduction 2 The Parallel Block Matrix Computing Algorithm and Linear Array 3 Multi-precision Floating-Point Matrix Computing Based on a Unified Structure 3.1 Two Parallel Single-Precision Floating-Point Matrix Computing 3.2 Four Parallel Half-Precision Floating-Point Matrix Computing 3.3 Multi-precision Floating-Point Multiply-Accumulate (Subtract) Functional Component 3.4 Implementation for Multi-precision MCA Unit Core and MCA System 4 Experimental Results and Discussions 4.1 Synthesis Results 4.2 Performance Analysis 4.3 Discussion 5 Conclusions References A Configurable Hardware Architecture for Runtime Application of Network Calculus 1 Introduction 2 Related Work 3 Configurable Hardware Architecture 3.1 Micro-architecture for Function f Ø f (Arrival Curve) 3.2 Sampling-Based Micro-architecture for Arrival Curve 3.3 Micro-architecture for Function f Ø g 3.4 Micro-architecture for Function f⊗g 3.5 Unified Micro-architecture with Function Configuration 4 FPGA Implementation and Evaluation 4.1 Performance Optimization 4.2 Scalability and Overhead 4.3 Comparison with Software Implementation 5 System Prototype and Case Study 6 Conclusion References FEB3D: An Efficient FPGA-Accelerated Compression Framework for Microscopy Images 1 Introduction 2 Background 2.1 B3D Algorithm 2.2 FPGA and OpenCL 3 Design and Implementation 3.1 The Co-design Framework 3.2 Hardware Optimized Design 3.3 Further Optimizations 4 Experiment and Evaluation 4.1 Experimental Setup 4.2 Results and Analysis 5 Related Work 6 Conclusion References NUMA-Aware Optimization of Sparse Matrix-Vector Multiplication on ARMv8-Based Many-Core Architectures 1 Introduction 2 Background 2.1 Parallel Sparse Matrix-Vector Multiplication 2.2 NUMA Architecture of the Phytium 2000+ Processor 2.3 Hypergraph Partitioning 3 NUMA-Aware SpMV Alogrithm 4 Performance and Analysis 4.1 Experimental Setup and Dataset 4.2 NUMA-Aware SpMV Performance 4.3 Preprocessing Overhead (Hypergraph Partitioning Runtime) 5 Related Work 6 Conclusions References CARAM: A Content-Aware Hybrid PCM/DRAM Main Memory System Framework 1 Introduction 2 System Design of CARAM 2.1 The Overview of System Architecture 2.2 Line Deduplication Processing 3 Evaluation 4 Conclusions References Big Data and Cloud Optimization of RDMA-Based HDFS Data Distribution Mechanism 1 Introduction 2 Related Work 3 Background and Motivation 3.1 HDFS Introduction 3.2 The Writing Process of Original HDFS 3.3 RDMA Introduction 4 The Design and Implementation of JUCX HDFS 5 Evaluation 5.1 Experimental Setup 5.2 Single Replica 5.3 Multiple Replicas 6 Conclusion References Reducing the Time of Live Container Migration in a Workflow 1 Introduction 2 Background and Motivation 2.1 Live Migration 2.2 Motivation 3 Overview of Architecture 4 Building Migration Engine 4.1 Migration Engine 4.2 Modification to Docker 4.3 Case Study: Analysis of Container Migration Time Reduction 5 Optimizing Migration Tool 5.1 Workflow Migration 6 Experiment and Evaluation 6.1 Methodology 6.2 Optimization of Container Startup 6.3 Test on Workflow Migration 7 Related Work 8 Conclusion References RDMA-Based Apache Storm for High-Performance Stream Data Processing 1 Introduction 2 Background and Motivation 2.1 Parallel Processing of Data Streams 2.2 Message Processing Structure 2.3 InfiniBand and RDMA 2.4 Analysis 3 Design 3.1 Messaging Transport Layer and Basic Model 3.2 JXIO-based Implementation 3.3 DiSNI-Based Implementation 4 Performance Evaluation 4.1 Experimental Setup 4.2 The Effect of Message Size on Performance 4.3 The Impact of Distributed Scale on Performance 5 Conclusion 6 Related Work References Payment Behavior Prediction and Statistical Analysis for Shared Parking Lots 1 Introduction 2 Background and Motivation 3 Payment Behavior Prediction and Statistical Analysis 4 Experiment 5 Conclusion References Edge Computing Location-Based Service Recommendation for Cold-Start in Mobile Edge Computing 1 Introduction 2 Related Work 3 Motivation 4 Multidimensional Inverse Similarity Recommendation Algorithm Based on CF 4.1 QoS Prediction 4.2 Service Recommendation 4.3 Algorithm Design 5 Experiment 5.1 Experiment Environment and Datasets 5.2 Experimental Design 5.3 Experiment Results 6 Conclusion References An Adaptive Delay-Limited Offloading Scheme Based on Multi-round Auction Model for Mobile Edge Computing 1 Introduction 2 Related Work 2.1 Offloading Scheme with the Goal of Reducing Latency 2.2 Offloading Scheme with the Goal of Reducing Energy Consumption 2.3 Offloading Scheme with the Trade-Off Between Latency and Energy Consumption 3 Adaptive Delay-Limited Offloading Scheme Based on Multi-round Auction Model 3.1 Scenario Description 3.2 Calculation Model 3.3 Multi-round Auction Model 3.4 Adaptive Delay-Limited Mechanism 4 Simulation Results and Analysis 5 Conclusion References An Efficient Data Transmission Strategy for Edge-Computing-Based Opportunistic Social Networks 1 Introduction 2 System Model 2.1 Community Definition 2.2 Data Transmission Model 3 Simulation and Analysis 4 Conclusion References Emering Shadow Data: A Method to Optimize Incremental Synchronization in Data Center 1 Introduction 2 Background and Related Work 2.1 Incremental Synchronization 2.2 Related Work 3 The Design of Shadow Data 3.1 What Makes the CPU Load in the Data Center 3.2 Why Shadow Data 3.3 What Is Shadow Data 4 Evaluation 4.1 Experiment Setup 4.2 CPU Load of the Data Center 4.3 Sacrifice of Disk on Local Node 5 Conclusion References DROAllocator: A Dynamic Resource-Aware Operator Allocation Framework in Distributed Streaming Processing 1 Introduction 2 Framework 2.1 Overview 2.2 Node Resources Predictor 2.3 Node Clusterer 2.4 Resource Aware Scheduler 3 Experiments 3.1 Datasets and Settings 3.2 Metrics 3.3 Results 4 Conclusion References A Medical Support System for Prostate Cancer Based on Ensemble Method in Developing Countries 1 Introduction 2 Design of Medical Support System 2.1 Overall Requirements and Framework of the System 2.2 Detailed Description of the Medical Support System 3 Experiment 3.1 Dataset and Models' Training 3.2 Analysis of the Results of Experiments 3.3 Relevant Analysis Based on the System 4 Conclusion References The Entry-Extensible Cuckoo Filter 1 Introduction 2 Related Work 2.1 Static Structures 2.2 Dynamic Structures 3 Entry-Extensible Cuckoo Filter 3.1 Overview 3.2 Entry-Level Extension 3.3 Operations of E2CF 3.4 Fine-Grained Splitting 4 Analysis of E2CF 4.1 Query Performance 4.2 False Positive Rate 5 Performance 5.1 Experiment Setups 5.2 Results 6 Conclusion References Monitoring Memory Behaviors and Mitigating NUMA Drawbacks on Tiered NVM Systems 1 Introduction 2 The Art of Our Design-3pt 2.1 SysMon-N - Sampling Memory Systems with NVM 2.2 N-Policy - for NVM 3 Effectiveness of N-Policy on Bandwidth and Latency References Network TPL: A Novel Analysis and Optimization Model for RDMA P2P Communication 1 Introduction 2 Background 2.1 Low-Level Details Related to RDMA P2P Communication Performance 2.2 Optimization Method Focusing on Hardware View and Single Message Transmission Process 3 TPL: Analysis and Optimization Model of RDMA P2P Communication Performance 3.1 Transmission Efficiency of a Single Message 3.2 Parallelism of WQE Submission and Handling 3.3 Load Balance Between CPU and NIC 3.4 Relation Between Three Dimensions 4 Optimization Algorithm Design Based on TPL 4.1 TPL-Based Systematic Analysis and Single-Dimensional Optimization 4.2 TPL-Based Multi-dimensional Optimization 5 Evaluation 5.1 Testing and Analysis of Dynamic Signal Period 5.2 Throughput Test of ORCL-Advanced 6 Related Work 7 Conclusion References Connectivity and Routing Algorithm of the Data Center Network HSDC 1 Introduction 2 Preliminaries 2.1 Terminology and Notation 2.2 Topological Structure of Hn 3 Connectivity and Tightly Super Connectivity of Hn 4 Shortest Routing Algorithm 5 Conclusions References CCRP: Converging Credit-Based and Reactive Protocols in Datacenters 1 Introduction 2 Background and Motivation 2.1 DCN Needs the Credit-Based Protocols 2.2 The Challenges of Deploying Credit-Based Protocols in DCNs 3 CCRP Design 3.1 Basic Idea 3.2 ECN-Based Feedback Control 3.3 Parameter Choice 3.4 Endhost and Switch Mechanism 4 Evaluation 5 Related Work 6 Conclusion References Storage CompressedCache: Enabling Storage Compression on Neuromorphic Processor for Liquid State Machine 1 Introduction 2 Background and Motivation 2.1 Liquid State Machine 2.2 Compressed Sparse Storage Formats 2.3 Motivation 3 Compressed Sparse Set Associative Cache 3.1 Details of the Method 3.2 Metadata Overhead 4 CSSAC-Improved Architecture 4.1 Hardware Neuron 5 Experiment Setup 5.1 Simulation Infrastructure 5.2 Hardware Implementation 5.3 Datasets 5.4 Measurements 6 Evaluation 6.1 Discard Ratio Sensitivity Analysis on Number of Sets 6.2 Compression Effect Sensitivity Analysis on Weight Data Width 6.3 Power Consumption Evaluation 6.4 Performance Evaluation 7 Related Work 8 Conclusion References ODCP: Optimizing Data Caching and Placement in Distributed File System Using Erasure Coding 1 Introduction 2 Background and Motivation 2.1 Erasure Coding 2.2 Data Cache Allocation 2.3 File Read Probability 2.4 Motivation 3 Design and Implementation 3.1 Data Cache Allocation Using Simulated Annealing 3.2 Building File Read Delay Model 3.3 Formulating the Optimization Problem 3.4 Solving the Convex Optimization Problem 4 Evaluation 4.1 Experimental Setup 4.2 File Read Latency Optimization 4.3 Parameter Sensitivity Analysis 5 Related Work 6 Conclusion References Towards Optimizing Deduplication on Persistent Memory 1 Introduction 2 Background and Related Work 2.1 Data Deduplication 2.2 Persistent Memory 3 Motivation 4 Modeling and Analysis 5 Design Choices 6 Experimental Results 6.1 Experiment Settings 6.2 Overall Performance 6.3 Quantitative Analysis 6.4 Bloom Filter Size 7 Conclusion References Author Index