CXL.mem and the resulting memory pool are promising and gaining great attention. Unlike local memory, CXL DIMMs stay at the I/O subsystem, whose inferior performance can easily impact the processor pipeline and memory subsystem, yielding performance interference, hardware contention, obscure behaviors, and underutilized communication and computing resources. However, our community lacks a tool to understand and profile the CXL.mem protocol execution end-to-end between CPU and remote DIMM. This paper fills the gap by designing and implementing PathFinder1, a systematic, informative, and lightweight CXL.mem profiler. PathFinder leverages the capabilities of existing hardware performance monitors (PMUs) and dissects the CXL.mem protocol at adequate granularities. Our key idea is to view the server processor and its chipset as a multi-stage Clos network, equip each architectural module with a PMU-based telemetry engine, track different CXL.mem paths, and apply conventional traffic analysis techniques. PathFinder performs snapshot-based path-driven profiling and introduces four techniques, i.e., path construction, stall cycle breakdown, interference analyzer, and cross-snapshot analysis. We build PathFinder atop Linux Perf and apply it to seven case studies.

Graph Neural Network(GNN) has achieved remarkable progress in the field of graph representation learning. The most prominent characteristic, propagating features along the edges, degrades its performance in most heterophilic graphs. Certain researches make attempts to construct KNN graph to improve the graph homophily. However, there is no prior knowledge to choose proper K and they may suffer from the problem of Inconsistent Similarity Distribution (ISD). To accommodate this issue, we propose Probability Graph Complementation Contrastive Learning (PGCCL) which adaptively constructs the complementation graph. We employ Beta Mixture Model(BMM) to distinguish intra-class similarity and inter-class similarity. Based on the posterior probability, we construct Probability Complementation Graphs to form contrastive views. The contrastive learning prompts the model to preserve complementary information for each node from different views. By combining original graph embedding and complementary graph embedding, the final embedding is able to capture rich semantics in the finetuning stage. At last, comprehensive experimental results on 20 datasets including homophilic and heterophilic graphs firmly verify the effectiveness of our algorithm as well as the quality of probability complementation graph compared with other state-of-the-art methods.

Semi-Supervised Graph Learning (SSGL) aims to predict massive unknown labels based on a subset of known labels within a graph. Recently, graph neural network, one of the most popular SSGL approaches, has garnered considerable research interest and achieved remarkable progress. However, many of these methods perform unsatisfactorily with limited labeled data. Graph contrastive learning (GCL), which utilizes unlabeled data to generate supervision, partially addresses this issue but does not fully exploit label information. To address this challenge, we propose SSGL algorithm, Semi-supervised Graph Contrastive Learning with Confidence Propagation Algorithm (SGCL). SGCL comprises two stages of contrastive learning. In the first stage, we employ unsupervised contrastive learning to initialize the model with graph augmentation. In the second stage, in order to fully leverage known labels and graph structure, we incorporate supervised contrastive learning which utilizes supervision signals obtained from confidence propagation algorithm. By combining supervised contrastive learning and unsupervised contrastive learning, the embedding quality and the classification accuracy can be further enhanced. At last, comprehensive experiments demonstrate that SGCL outperforms the best baseline method by an average of 2.23% across six datasets, highlighting the effectiveness of our approach.

In vertical federated learning (VFL), multiple participants can collaborate in training a model with distributed data features and labels managed by one of them. The cooperation provides opportunities for a malicious participant to conduct a backdoor attack. However, the attack is challenging when the adversary does not own labels with the mitigation of other participants. In this paper, we discover that an adversary can exploit local latent representations output in the inference stage to inject a backdoor in VFL, even without access to labels. With little auxiliary labeled data, the adversary fine-tunes its bottom model to make it output specific latent representation for backdoor input instances, which induces the federated model to predict the attacker-specified label regardless of benign participants. Our experiments show that the proposed attack can achieve a high attack success rate with little loss of main task accuracy and outperform existing backdoor attacks. We also explore possible defenses against the attack. Our research demonstrates the potential security threat to VFL.

With DNN turning into the backbone of AI cloud services and propelling the emergence of INFerence-as-a-Service (INFaaS), DNN-specific accelerators have become the indispensable components of cloud inference systems. Due to the conservative “one-task-at-a-time” working mode and deadline blindness of those accelerators, implementing multi-tenancy that aims to improve the cost-effectiveness and meet SLA requirements is intractable. Recent studies including the temporal and spatial approaches, employ manifold scheduling mechanisms and sophisticated architecture innovations to address the challenge. However, these researches either still neglect the deadline awareness or render inevitable and expensive hardware overheads such as switches and storage. In this paper, we present Cooperative and Deadline-aware Multi-Systolic-Array scheduling (CD-MSA), a low-cost solution for the cloud inference that utilizes the real time mechanism and task-level parallelism to enable efficient multi-tenancy. Based on our preemptive multi-systolic-array accelerator architecture supporting the simultaneous task co-location, we first construct a fine-grained DNN execution model to lay the groundwork for the lightweight preemption. Second, we design a cooperative, deadline- and laxity-aware scheduler in conjunction with an efficient schedulability test method for better QoS guarantee without introducing additional hardware cost. Finally, to further promote the overall throughput, we propose dynamic task fusion , a software approach that fuses different tasks into the logically “multi-threading” tasks at runtime. We compare CD-MSA with several state-of-the-art researches across three multi-DNN workloads. The evaluation results show CD-MSA improves the latency-bounded throughput, SLA satisfaction rate and weighted system throughput by up to 62%, 63% and 27%, respectively.

随着数据孤岛现象的出现和个人隐私保护的重视,集中学习的应用模式受到制约,而联邦学习作为一个分布式机器学习框架,可以在不泄露用户数据的前提下完成模型训练,从诞生之初就备受关注.伴随着联邦学习应用的推广,其安全性和隐私保护能力也开始受到质疑.对近年来国内外学者在联邦学习模型安全与隐私的研究成果进行了系统总结与分析.首先,介绍联邦学习的背景知识,明确其定义和工作流程,并分析存在的脆弱点.其次,分别对联邦学习存在的安全威胁和隐私风险进行系统分析和对比,并归纳总结现有的防护手段.最后,展望未来的研究挑战和方向.

The recent emergence of low-power high-throughput programmable storage platforms-SmartNIC JBOF (just-a-bunch-of-flash)-motivates us to rethink the cluster architecture and system stack for energy-efficient large-scale data-intensive workloads. Unlike conventional systems that use an array of server JBOFs or embedded storage nodes, the introduction of SmartNIC JBOFs has drastically changed the cluster compute, memory, and I/O configurations. Such an extremely imbalanced architecture makes prior system design philosophies and techniques either ineffective or invalid. This paper presents LEED , a distributed, replicated, and persistent key-value store over an array of SmartNIC JBOFs. Our key ideas to tackle the unique challenges induced by a SmartNIC JBOF are: trading excessive I/O bandwidth for scarce SmartNIC core computing cycles and memory capacity; making scheduling decisions as early as possible to streamline the request execution flow. LEED systematically revamps the software stack and proposes techniques across per-SSD, intra-JBOF, and inter-JBOF levels. Our prototyped system based on Broadcom Stingray outperforms existing solutions that use beefy server JBOFs and wimpy embedded storage nodes by 4.2×/3.8× and 17.5×/19.1× in terms of requests per Joule for 256B/1KB key-value objects.

SmartNICs have become an indispensable communication fabric and computing substrate in today’s data centers and enterprise clusters, providing in-network computing capabilities for traversed packets and benefiting a range of applications across the system stack. Building an efficient SmartNIC-assisted solution is generally non-trivial and tedious as it requires programmers to understand the SmartNIC architecture, refactor application logic to match the device’s capabilities and limitations, and correlate an application execution with traffic characteristics. A high-level SmartNIC performance model can decouple the underlying SmartNIC hardware device from its offloaded software implementations and execution contexts, thereby drastically simplifying and facilitating the development process. However, prior architectural models can hardly be applied due to their limited capabilities in dissecting the SmartNIC-offloaded program’s complexity, capturing the nondeterministic overlapping between computation and I/O, and perceiving diverse traffic profiles. This paper presents the LogNIC model that systematically analyzes the performance characteristics of a SmartNIC-offloaded program. Unlike conventional execution flow-based modeling, LogNIC employs a packet-centric approach that examines SmartNIC execution based on how packets traverse heterogeneous computing domains, on-/off-chip interconnects, and memory subsystems. It abstracts away the low-level device details, represents a deployed program as an execution graph, retains a handful of configurable parameters, and generates latency/throughput estimation for a given traffic profile. It further exposes a couple of extensions to handle multi-tenancy, traffic interleaving, and accelerator peculiarity. We demonstrate the LogNIC model’s capabilities using both commodity SmartNICs and an academic prototype under five application scenarios. Our evaluations show that LogNIC can estimate performance bounds, explore software optimization strategies, and provide guidelines for new hardware designs.

This article proposes an innovative ground traffic management approach that utilizes the mobility, flexibility, and collaborative capabilities of multiple Unmanned Aerial Vehicles (UAVs). The objective is to enhance the navigation and driving experience of self-driving vehicles by employing UAVs to cover blind areas that cannot be observed by ground monitoring equipment, thereby facilitating the avoidance of congested routes. This study focuses on determining the optimal number of UAVs and enhancing the scheduling strategy and task assignment method by introducing novel UAV communication collaboration techniques. The research faces challenges such as a limited quantity of UAVs, rapid response and feedback, variations in traffic conditions and tasks, and system complexity. Furthermore, the difficulty of communication collaboration is exacerbated by the incomplete connectivity of UAV networks. To address these challenges, this article introduces a method that utilizes intention information as the message for communication between UAVs, thereby enhancing their collaborative capabilities. Using the Multi-Agent Reinforcement Learning (MARL) neural network, this method generates intention information and utilizes the intention information generated by other UAVs to optimize the flight control and traffic monitoring task allocation strategy. This method effectively addresses the challenge of UAV collaboration in efficiently managing a large volume of stochastic traffic monitoring tasks. Experimental results demonstrate that the proposed approach significantly improves vehicle navigation accuracy and achieves a more balanced distribution of workload among multiple UAVs. Moreover, by sharing road conditions detected in the cloud, it reduces operational costs and opens up possibilities for the practical implementation of UAVs in traffic management.

With the popularity of IoT (Internet of Things) applications, edge computing has received lots of attention. To meet data privacy protection requirements of edge nodes and cope with their unbalanced data distribution, federated learning (FL), a distributed learning framework, is widely used in intelligent edge computing applications. However, recent studies have shown that FL still suffers from privacy leakage problems, including membership inference, data reconstruction, etc. However, these studies mainly focus on the feature information of private data. In this paper, we concern the user-level label privacy in FL. We propose LDIA, a label distribution inference attack against FL in edge computing, exploring the possibility that an honest but curious cloud server can infer the proportions of samples per label in the edge user's private data. LDIA is inspired by the observation that parameter changes in the output layer of a model can reflect the label distribution of training data. We use a neural network to learn individual features of the output layer updates over different label distributions, and then perform inference from local models uploaded by users. Our comprehensive evaluation shows that LDIA is effective on various datasets in different settings, demonstrating the severe privacy leakage in FL-based edge computing.

Dynamic Graph Neural Networks (DGNNs) have been widely applied in various real-life applications, such as link prediction and pandemic forecast, to capture both static structural information and temporal characteristics from dynamic graphs. Combining both time-dependent and -independent components, DGNNs manifest substantial parallel computation and data reuse potentials, but suffer from severe memory access inefficiency and data transfer overhead under the canonical one-graph-at-a-time training pattern. To tackle these challenges, we propose PiPAD, a Pipelined and PArallel DGNN training framework for the end-to-end performance optimization on GPUs. From both algorithm and runtime level, PiPAD holistically reconstructs the overall training paradigm from the data organization to computation manner. Capable of processing multiple graph snapshots in parallel, PiPAD eliminates unnecessary data transmission and alleviates memory access inefficiency to improve the overall performance. Our evaluation across various datasets shows PiPAD achieves 1.22 × --9.57× speedup over the state-of-the-art DGNN frameworks on three representative models.

Telecommunication network plays an important role in our daily life. Fault detection and alarm root cause analysis are the keys to ensure the normal operation of the network. To reduce the burden on operators, numerous methods are employed to analyse root cause of faults. However, there still remain a large amount of non-essential or transient alarms after root cause analysis. A simple Rule-based method may help ease the problems. But it needs prior expert knowledge and the diversity of alarm pattern makes the rules redundant and complicated. Moreover, it cannot accurately cover all true faults and need manual methods as complement. In this work, we propose Alarm Propagation Graph Neural Network(APGNN), a novel data-driven propagation-based root cause analysis and fault detection approach.It first associates alarms and extracts root-derived graph based on Bayesian Network. Then it constructs alarm propagation graphs(APG). We refine the repair orders to obtain actual fault information. At last, Graph Neural Network is used to extract features and learn the mapping from APG to the true fault. Our method not only detects the true fault from large volume of original alarms, but also analyses the root cause alarms. We evaluate our approach both on the offline and online environment of the real-world IP Radio Access Network. Experiments show that our model outperforms the state-of-art approach by 4.6% in F1-score on average.

Federated learning (FL) is vulnerable to membership inference attacks even it is designed to protect users’ data during model training, as model parameters remember the information of training data. However, existing inference attacks against FL perform poorly in multi-participant scenarios. We propose CS-MIA, a novel membership inference based on prediction confidence series, posing a more critical privacy threat to FL. The inspirations of CS-MIA are the different prediction confidence of a model on training and testing data, and multiple versions of target models over rounds during FL. We use a neural network to learn individual features of confidence series on training and testing data for subsequent membership inference. We design inference algorithms for both local and global adversaries in FL. And we also design an active attack for global adversaries to extract more information. Our confidence-series-based membership inference outperforms most state-of-the-art attacks on various datasets in different scenarios, demonstrating the severe privacy leakage in FL.

The use of multi-agent reinforcement learning methods (MARL) in mobile edge computing (MEC) environments enables multiple unmanned aerial vehicles (multi-UAV) to intelligently provide relay or computational offloading services to mission targets. UAV's observation range and communication methods between UAVs have a significant impact on multi-UAV collaboration strategy. For this purpose, we study the multi-UAV observation range dynamic control method and the optimal inter-UAV communication method. Our approach is to design a multi-UAV joint observation, communication, policy, and service collaboration protocol and study the optimization method of the protocol. We propose an expert-guided deep reinforcement learning framework to optimize this protocol. Each UAV's optimal radar observation range and inter-UAV communication method are learned using an information entropy value decomposition method. Through our observation and communication method, multi-UAV are able to obtain the most valuable information. Experiments demonstrate that our method can improve MEC's service coverage by 9.38%-21.88% compared to the classical MARL algorithm. Our method improves the radar observation efficiency and communication efficiency by 3.05%-38.9% and 8.55%-22.03%, respectively. The results show that this method improves multi-UAV energy utilization.

With the rapid development of wireless network technology and the demand for interconnection of mobile edge computing, the integration of heterogeneous wireless networks is becoming more and more urgent. At the same time, the increasing scarcity of spectrum resources also requires heterogeneous networks to effectively improve the utilization of spectrum resources without affecting the use of different licensed spectrum resources. This paper proposed a multi-channel millimeter-wave (mmWave) MAC protocol for mobile self-organizing heterogeneous networks. The mmWave Mac uses a bulk access transmission scheme to improve throughput efficiency and reduce latency for heterogeneous networks. In addition, cognitive radio technology is also introduced into the mmWave Mac to enhance the utilization of spectrum resources. Finally, by comparing the simulation results with other protocol, it is verified that the mmWave MAC protocol has good performance in terms of delay, and throughput efficiency.

In delay-tolerant networks (DTNs), intermittent network connectivity and lack of global system information pose serious challenges to achieve effective data forwarding. Most state-of-the-art DTN routing algorithms are based on hill-climbing heuristics in order to select the best available next hop to achieve satisfactory network throughput and routing efficiency. An adverse consequence of this approach is that a small subset of good users take on most of the forwarding tasks. This can quickly deplete scarce resources (e.g. storage, battery, etc.) in heavily utilized devices which degrades the network reliability. A system with a significant amount of traffic carried by a small number of users is not robust to denial of service attacks and random failures. To overcome these deficiencies, this paper proposes a new routing algorithm, DTN-Balance, that takes the forwarding capacity and forwarding queue of the relay nodes into account to achieve a better load distribution in the network. For this, we defined a new routing metric called message forwarding utility combining nodal available bandwidth and forwarding workload. Applying small world theory, we impose an upper bound on the end-to-end hop count that results in a sharp increase in routing efficiency. Queued messages in a forwarding node are arranged by DTN-Balance based on message dropping utility metric for a more intelligent decision in the case of a message drop. The performance of our method is compared with that of the existing algorithms by simulations on real DTN traces. The results show that our algorithm provides outstanding forward efficiency at the expense of a small drop in the throughput.

Accelerators, like GPUs, have become a trend to deliver future performance desire, and sharing the same virtual memory space between CPUs and GPUs is increasingly adopted to simplify programming. However, address translation, which is the key factor of virtual memory, is becoming the bottleneck of performance for GPUs. In GPUs, a single TLB miss can stall hundreds of threads due to the SIMT execute model, degrading performance dramatically. Through real system analysis, we observe that the OS shows an advanced contiguity (e.g., hundreds of contiguous pages), and more large memory regions with advanced contiguity tend to be allocated with the increase of working sets. Leveraging the observation, we propose MESC to improve the translation efficiency for GPUs. The key idea of MESC is to divide each large page frame (2MB size) in virtual memory space into memory subregions with fixed size (i.e., 64 4KB pages), and store the contiguity information of subregions and large page frames in L2PTEs. With MESC, address translations of up to 512 pages can be coalesced into single TLB entry, without the needs of changing memory allocation policy (i.e., demand paging) and the support of large pages. In the experimental results, MESC achieves 77.2% performance improvement and 76.4% reduction in dynamic translation energy for translation-sensitive workloads.

Due to the broad attack surface and the lack of runtime protection, potential safety and security threats hinder the real-life adoption of autonomous vehicles. Although efforts have been made to mitigate some specific attacks, there are few works on the protection of the autonomous driving system, i.e., the control software system performing such as perception, decision making, and motion tracking. This article presents a decentralized self-protection framework called Guardauto to protect the autonomous driving system against runtime threats. First, Guardauto proposes an isolation model to decouple the autonomous driving system and isolate its components with a set of partitions. Second, Guardauto provides self-protection mechanisms for each target component, which combines different methods to monitor the target execution and plan adaption actions accordingly. Third, Guardauto provides cooperation among local self-protection mechanisms to identify the root-cause component in the case of cascading failures affecting multiple components. A prototype has been implemented and evaluated on the open-source autonomous driving system Autoware. Results show that Guardauto could effectively mitigate runtime failures and attacks, and protect the control system with acceptable performance overhead.

Unmanned aerial vehicle (UAV) is one of the preferred tools for coverage detection missions, because of its maneuverability and flexibility. It is challenging for the UAV to decide a track by itself in a complex geometrical environment. This paper presents a UAV intelligent navigation method based on deep reinforcement learning (DRL). We propose using geographic information systems (GIS) as the DRL training environment to overcome the inconsistency between the training environment and the test environment. We creatively save the flight path in the form of an image. The combination of the knowledge-based Monte Carlo tree search method and local search method can not only effectively avoid falling into local search, but also ensure learning the optimal search direction under the limitation of computing power. Experiments show that the trained UAV can find an excellent flight path by intelligent navigation, and able to make effective flight decisions in a complex geometrical environment.