The increasing complexity of deep learning models and growing GPU memory demands exacerbate memory management challenges. Unified Memory (UM) offers a transparent solution by enabling models to exceed hardware memory limits, yet it suffers from performance bottlenecks such as prefetching delays and frequent page faults during training. To address this, we propose a Temporal History-based Dynamic Prefetching (THDP) approach, which leverages kernel execution history to predict memory access patterns. THDP dynamically adjusts prefetching windows and employs a reference-countbased eviction strategy to optimize data migration. Evaluations across diverse deep learning models demonstrate that THDP significantly outperforms baselines, achieving an average speedup of$2.04 \times$and reducing page faults by over 90%.

In cloud storage services, burst I/O workloads from data analytics and artificial intelligence/machine learning (AI/ML) applications present significant challenges to Quality of Service (QoS) management. Existing scheduling models like dmClock ensure fair and stable I/O bandwidth allocation in typical scenarios. However, they falter under frequent burst traffic, leading to lower resource utilization and higher task latency. To address this, we propose BAQoS, a Burst I/O Aware Quality of Service Optimization for Cloud Storage Service. BAQoS employs refined request classification, a burst-aware hierarchical scheduling algorithm, and a high-performance scheduler architecture (HPSA). These features enable dynamic resource allocation and efficient scheduling for both burst and regular requests. Experiments show that BAQoS markedly enhances performance under burst workloads, accelerating burst request processing by up to 7.09 and improving overall system performance by 48.86%. Furthermore, BAQoS ensures superior performance for non-burst users, achieving a 51.15% performance boost for high-priority users, a 5.34% increase in system throughput, and an over 50% reduction in IOPS standard deviation among same-priority users.

The Digital Twin (DT) technique employs virtual models to accurately represent physical entities and their dynamic behaviors, enabling applications in prediction, optimization, and real-time monitoring. When integrated with deep learning (DL) and offered as a service, DTs significantly enhance system intelligence and operational efficiency. However, resource constraints in edge environments pose significant challenges for GPU resource scheduling, particularly under concurrent execution of DLI tasks for DT services. To address these challenges, this paper introduces an SLO-aware scheduling framework designed to optimize DL inference for DT-enabled intelligent edge systems, which combines accurate task processing latency prediction with a dynamic synchronization strategy. The proposed strategy optimizes GPU resource allocation to meet task-specific performance objectives, thereby improving system productivity and efficiency. Comprehensive evaluations show that our method significantly reduces SLO violations by 31.2% to 90.1% and JCT by 16.2% to 80.3% compared to baseline methods, demonstrating its effectiveness in resource-limited edge computing environments under high workload scenarios.

Caching recommended contents at the network edge can effectively alleviate the traffic pressure of the backbone network and significantly improve user experience. However, highly personalized and precise recommendations often rely on leveraging more user request records, raising serious privacy concerns. Existing recommendation-aware edge caching mechanisms typically apply a fixed level of privacy protection, without considering the personalized privacy of users. This one-size-fits-all approach often introduces significant noise, adversely impacting cache hit ratio (CHR). In this work, we propose a differential privacy-based edge caching framework supporting personalized privacy-preserving to address these challenges. We formulate a CHR maximization problem under personalized privacy constraints and reveal the NP-completeness of the problem with a rigorous mathematical proof. Subsequently, we mathematically model the relationship between personalized privacy and user preference distortion, analyzing its impact on recommendations and user requests. To solve it, we introduce an efficient heuristic algorithm named the Backhaul Traffic-Aware Caching Algorithm. This algorithm utilizes backhaul traffic as a feedback signal to make accurate caching decisions, enabling adaptive optimization of caching decisions by perceiving the impact of noise and low-quality recommendations. Extensive experiments on two typical real-world datasets validate the effectiveness of our framework, demonstrating its ability to enhance privacy protection while simultaneously improving CHR.

Distributed applications are increasingly deployed at the edge to provide low-latency user access. The hierarchical and localized distribution of edge nodes presents challenges for traditional consensus protocols. The push-based replication method, where the leader pushes updates to followers, can become a performance bottleneck in bandwidth-constrained edge environments. Furthermore, the variation in resources and processing capabilities among edge nodes makes it difficult for this replication method to adapt to the specific conditions of each node, thereby affecting system performance. This paper proposes a new consensus protocol, PPaxos, which employs a pull-based replication method. In this approach, follower nodes within a group proactively pull data from relay nodes, and the relay nodes pull data from the leader node. We have designed an adaptive load-aware pull strategy that allows each node to dynamically adjust its pull frequency based on its own state and to dynamically select relay nodes to reduce performance fluctuations and instability factors. Additionally, a cross-group multi-link communication mechanism ensures that the system remains stable even during relay node failures. Experimental results indicate that PPaxos outperforms other protocols in terms of throughput and latency. Specifically, under conditions of limited bandwidth in edge networks, PPaxos achieves approximately a ${2 9 . 6 \%}$ improvement in throughput and a ${3 4 . 8 8 \%}$ reduction in latency.

The rapid growth of medical data in cloud servers offers a rich resource for medical research and analysis. However, the sensitive nature of this data necessitates robust access controls to ensure privacy and confidentiality. Balancing the utilization of valuable medical information with the protection of personal data privacy poses a significant challenge. To address this challenge, this paper proposes a Hierarchical Retrieval scheme with Bilateral security based on Attribute encryption (BAHR) to facilitate multi-level access control. Based on an attribute-based encryption architecture, the hierarchical access control categorizes users into distinct access groups, allowing for precise data retrieval while maintaining encryption integrity at specified levels. Furthermore, the proposed scheme ensures bilateral security, providing robust protection of patients’ private data, database, and user access patterns within untrusted domains. Through security analysis and experimental evaluation, BA-HR demonstrates good performance regarding retrieval time, decryption time and storage space.

Stateful serverless systems commonly adopt an architectural paradigm characterized by compute and storage separation within cloud data centers. Nevertheless, guaranteeing prompt response for real-time tasks at the edge becomes challenging due to network overheads. This paper introduces a Low-Latency state management framework for real-time tasks in edge serverless systems called LoLa, which adaptively places states proximate to functions, thereby mitigating delays in accessing states within edge serverless systems. Our approach aims at mitigating network latency and optimizing resource utilization by co-locating functions and states, thereby enhancing the system’s overall efficiency. We introduce an adaptive strategy to coordinate the migration of states. It dynamically adjusts the positions of states based on historical data and real-time feedback. Additionally, we designed an in-memory state storage mechanism to facilitate low-latency access and implement a lightweight and fine-grained state management to ensure stored state consistency. Evaluation results showcase the efficacy of LoLa in reducing state read and write latency within edge serverless systems. Specifically, the average response latency is observed to decrease by 65.2% and 38.1% in the best and worst-case scenarios, respectively.

Erasure codes have been widely applied to in-memory key-value storage systems for high reliability and low redundancy. In distributed in-memory key-value storage systems, update operations are relatively frequent, especially the partial-stripe update, which makes data update more challenging. Recently, existing research has been based on appending logs to accelerate parity data write. However, its logs are stored on disks, which decreases the system performance significantly. Therefore, we propose a novel in-memory key-value storage architecture, DNVPL, which utilizes NVRAM to log parity data. Our main idea is to design an appending-only update scheme to tradeoff the memory cost and the update overhead. We implement DNVPL with an in-memory key-value storage prototype, called LogKV. We evaluate it with different workloads. The experiments show that our scheme achieves high update performance from different metrics. Our scheme can reduce update latency by up to 49% and save storage space by 48% compared to the state-of-the-art schemes.

Due to cloud security concerns, there is an increasing interest in information retrieval systems that can support private queries over public documents. It is desirable for oblivious document ranking and retrieval in public cloud at lower cost and faster speed without revealing query-related information. Currently, the term frequency-inverse document frequency (TF-IDF) and private information retrieval (PIR) techniques are used to solve this problem, but the encryption operation time is over dominant. Motivated by the observation of the sparsity of the TF-IDF matrix, we propose an efficient approach for oblivious document ranking and retrieval, called E-Coeus. It takes advantage of the high sparsity of the TF-IDF matrix to rearrange the matrix. Our method accelerates the speed of PIR inadvertently retrieving documents and reduces the user retrieval delay time. In a stand-alone experiment for a TF-IDF matrix of 1.2M rows and 64K columns with the sparsity of 10%, E-Coeus improves the document ranking and retrieval performance by 23% over the state-of-the-art approach, Coeus. With cluster of 64 machines, E-Coeus improves the performance by 34% over Coeus when the TF-IDF matrix sparsity is 30%.

Edge computing is the next-generation computing paradigm that brings the processing capability closer to the location where it is needed. 5G and beyond 5G aim to achieve substantial improvement for the performance of edge computing in terms of e.g. higher throughput and lower latency. Smart base stations are often attached with edge datacenters consisting of many edge servers equipped with computing and storage capabilities. These servers are used to execute offloaded tasks from edge equipment such as Internet of Things. It is important to have an efficient offloading algorithm that can guarantee specific service-level objectives (SLOs) by assigning tasks to appropriate edge servers. Traditional offloading schemes such as static and learning-based algorithms either have limited performance or result in high overhead for task assignment to servers. In this paper, we propose an efficient game-theoretical scheduling algorithm for offloaded tasks at edge datacenters. The core contribution of the algorithm is to design a public goods investment model for edge servers. Based on the model, we design a lightweight scheduling algorithm to reduce the average load of edge servers and enhance the stability of edge datacenter systems. Experimental results demonstrate the significant benefits of the proposed algorithm in reducing the response latency of tasks and balancing the workload of edge servers.

Distributed data services usually rely on consensus protocols like Paxos and Raft to provide fault-tolerance and data consistency across global and local-distributed data centers. Erasure coding replication has appealing storage and network cost saving compared with full copy replication, which helps consensus protocols achieve low latency, high fault tolerance, and high throughput for data access. Applying erasure coding in consensus protocols directly will degrade the liveness level when the number of failure servers reaches a certain level. To address the challenge, CRaft just stores full copy replication instead of erasure coding replication when the number of failed servers reaches a certain threshold. In such situation, CRaft will be downgraded sharply to the same storage and network costs as Raft. To overcome the shortcoming of CRaft, we propose a protocol, called HRaft, which can adapt the placement of data blocks in order to always have enough blocks to recover the stored value when servers fail. By replenishing some coded blocks in healthy servers instead of full copy replication, it can avoid switching to the full replication when a certain threshold on the number of failures is reached. We designed and implemented a key-value (KV) storage prototype to validate the proposed protocol and evaluate its performance. The experimental results show HRaft can significantly reduce storage and network costs and improve write performance while keeping the liveness level compared to CRaft.

Distributed data services usually rely on consensus protocols, such as Paxos and Raft, to provide fault-tolerance and data consistency across distributed data centers and even edge networks. In consensus protocols, erasure coded replication has appealing storage and network cost savings compared with full copy replication, which help achieve low latency, high fault-tolerance and high throughput. However, the liveness level will inevitably decrease when erasure codes are naively applied in consensus protocols. To keep the original liveness level, an existing protocol, called CRaft, switches from erasure coded replication to full copy replication when the number of failures exceeds a certain threshold. Such a solution, however, degrades system performance sharply. To tackle this problem, this work proposes a novel protocol called HRaft to enable graceful degradation on storage and network efficiency when failures happen. Without using full copy replication, it replenishes some coded blocks in healthy servers to reduce storage and network costs and to keep data consistency. The performance of the proposed protocol will be evaluated by deploving it into practical networks.

This work considers cache placement for ultra-dense fog radio access networks (F-RANs). In an F-RAN, the fog access points (F-APs) form overlapping clusters based on their geographical locations. A cluster of F-APs then acts as a distributed cache to cooperatively serve user requests. The fronthaul traffic minimization problem is formulated in information-theoretic terms. For the k-association networks, cache placement can be optimized by concatenating an MDS code with a repetition code. By repeating the same packet in some F-APs, multicasting over the fronthaul link can be done in cache placement, which saves energy and bandwidth. Such an idea is applied to both uncoded and coded caching schemes based on hypergraph coloring. Their associated optimization problems are shown to be NP-complete. For the uncoded case, a suboptimal algorithm is proposed, which carefully repeats the subfiles. For the coded case, a heuristic algorithm that minimizes the field size requirement of the MDS repetition scheme is proposed. Simulation results demonstrate the outstanding performance of the proposed uncoded and coded caching schemes during both the cache placement phase and content delivery phase in reducing fronthaul traffic load and energy consumption.

In fog radio access networks, the limited capacity of the fronthaul link is the bottleneck, which renders a high quality of service for video streaming difficult. To circumvent the problem, popular files can be cached in fog access points during off-peak hours. This work points out that beamforming in the access network can be exploited to reduce fronthaul traffic load by a joint design of cache placement scheme at the fog access points and index-coded transmission scheme over the fronthaul. Simulation results show that a percentage reduction of fronthaul traffic by more than 30% can be achieved.

In this paper, we present a learning based data deduplication algorithm, called LIPA, which uses the reinforcement learning framework to build an adaptive indexing structure. It is rather different from previous inline chunk-based deduplication methods to solve the chunk-lookup disk bottleneck problem for large-scale backup. In previous methods, a full chunk index or a sampled chunk index often is often required to identify duplicate chunks, which is a critical stage for data deduplication. The full chunk index is hard to fit in RAM and the sampled chunk index directly affects the deduplication ratio dependent on the sampling ratio. Our learning based method only requires little memory overheads to store the index but achieves the same or even better deduplication ratio than previous methods. In our method, after the data stream is broken into relatively large segments, one or more representative chunk fingerprints are chosen as the feature of a segment. An incoming segment may share the same feature with previous segments. Thus we use a key-value structure to record the relationship between features and segments: a feature maps to a fixed number of segments. We train the similarities of these segments to a feature represented as scores by the reinforcement learning method. For an incoming segment, our method adaptively prefetches a segment and the successive ones into cache by using multi-armed bandits model. Our experimental results show that our method significantly reduces memory overheads and achieves effective deduplication.

Cooperative caching using maximum distance separable (MDS) codes and repetition codes in ultra-dense networks is studied, with the objective of maximizing the code rate while ensuring that end users can restore the file from the associating small base stations (SBSs) without the use of the backhaul link. It is proved that MDS-coded caching is optimal in general. In contrast, repetition caching is optimal only for some special cases. Repetition caching is, in general, suboptimal, and the associated code rate maximization problem is shown to be NP-hard and a heuristic algorithm is designed to evaluate the potential coding gain in arbitrary 2-dimensional (2D) network. Simulation results show that MDS-coded caching can save about 40% storage space when compared with repetition caching, and this coding gain increases when the amount of overlapping between clusters increases.

Segment-tree (ST) based cost aggregation algorithm for stereo matching successfully integrates the information of segmentation with non-local cost aggregation framework. The tree structure which is generated by the segmentation strategy directly determines the final results for this kind of algorithms. However, the original strategy performs unreasonable due to its coarse performance and ignores to meet the disparity consistency assumption. To improve these weaknesses we propose a novel segmentation algorithm for constructing a more faithful ST with enhanced segmentation advantage according to a robust initial over-segmentation. Then we implement non-local cost aggregation framework on this new ST structure and obtain improved disparity maps. Performance evaluations on all 31 Middlebury stereo pairs show that the proposed algorithm outperforms than other five state-of-the-art aggregated based algorithms and also keeps time efficiency.

The instantiation of a maximally recoverable (MR) code is shown to be a special case of generic network coding. The defining condition of MR codes, called potential independence, is shown to be equivalent to maximal matching in bipartite graphs. Algorithms for MR instantiation are proposed and upper bounds on the required field size are derived.

Consider that a set of balls of n different colors are thrown into m bins with the assumption that the ball number of each color is constant and the number of balls in each bin is also constant. Our optimal goal is to find a feasible placement such that the distinct colors of remaining balls should be at least c after removing any k bins (k ≤ m) with the minimum number of balls. We present that the optimal colored bins in bins is equivalent to the optimization of Fractional Repetition (FR) codes in distributed storage systems. Here balls correspond to coded packets and bins correspond to storage nodes. This problem can be represented as biregualr graph and then deduced to the Zarankiewicz problem, which is a well-known extremal graph theoretic problem. We present the problem with the relation to combinatorial design theory, especially t-designs and propose the explicit construction algorithm for the optimization problem from t-designs. Some constructions of the optimized FR codes by 2-designs are analyzed to tolerate the desired k fault-tolerance with c = n - 1.

The minimum spanning tree (MST) based non-local cost aggregation algorithm performs well in accuracy and time efficiency. However, it can still be improved in two aspects. First, we propose a logarithmic transformation on matching cost function to improve the matching efficiency in texture less regions. The textureless neighbors can provide effective contributions in cost aggregation by the proposed monotone increasing function. Hence the algorithm can distinguish different pixels in textureless regions. Second, MST algorithm only utilizes color information in weight function while aggregating, which leads 3D cues missing. We introduce depth weight computed from the original MST algorithm into an edge weight function. With the proposed color-depth weight, we further iteratively rebuild the tree and obtain enhanced disparity map. Performance evaluations on 19 Middlebury stereo pairs and Microsoft stereo videos show that the proposed algorithm outperforms than other five state-of-the-art cost aggregation algorithms.

With the development of 3D technologies and devices, 3D model retrieval becomes a hot research topic where multi-view matching algorithms have demonstrated satisfying performance. However, exciting works overlook the common factors among objects in a single class, and they are time consuming in retrieval processing. In this paper, a class-statistics and pair-constraint model (CSPC) method is originally proposed for 3D model retrieval, which is composed of supervised class-based statistics model and pair-constraint object retrieval model. In our CSPC model, we firstly convert view-based distance measure into object-based distance measure without falling in performance, which will advance 3D model retrieval speed. Secondly, the generality of the distribution of each feature dimension in each class is computed to judge category information, and then we further adopt this distribution information to build class models. Finally, an object-based pairwise constraint is introduced on the base of the class-statistic measure, which can remove a lot of false alarm samples in retrieval. Experimental results on ETH, NTU-60, MVRED and PSB 3D datasets show that our method is fast, and its performance is also comparable with the-state-of-the-art algorithms.

To maximize the reliability of erasure coded objects in cloud storage, the optimal allocation is a challenging problem constrained with the node heterogeneities and erasure coding budgets. We model the optimal problem from a combinatorial view under multiple constraints and propose an efficient search algorithm to find the most reliable allocation. The experiments are evaluated and analyzed including the optimal reliability, redundancy reduction and allocation pattern.

To ensure high reliability and storage efficiency, erasure codes are preferred in storage systems. With the prevalent of distributed storage systems such as clouds storage, how to design a scalable and efficient erasure-resilient code is challenging. We propose a scalable binary linear code to keep up with data growth which has the following properties. Given the group size k and the code block length n, the proposed code corrects any two bit erasures among the n bits. The redundancy overhead of the code is 2/(k + 2), and each data bit affects exactly 2 parity bits. As results of these properties, if a data bit is changed or added, only two parity bits need to be updated; and the recovery of an erasured bit requires accessing at most k other bits and the recovery of two erasured bits requires at most 2k other bits. We give the construction algorithm by the order expansion of regular graphs; moreover, we optimize the failure resilience during the construction procedure. Compared with existing codes, our proposed code has notable benefits in storage scalability, redundancy overhead and I/O bandwidth. The deployment of the proposed code in distributed storage systems can be simple and practical.

In this paper, we study binary pattern erasure codes, i.e., binary codes that are resiliant to erasures from a family P of possible erasures. We give an algorithmic proof of the existence of a binary linear code with codewords of length n that is resiliant to erasures from P when P satisfies the properties: every pattern p ϵ P has size m and every letter in the alphabet occurs in at most c patterns. The density of the parity matrix is plays a important role in storage applications, so we also introduce a new low density code basing on graph theory.

Heterogeneity is the natural feature in distributed networks. Different from the traditional disk array, the amount of data allocated on heterogenous peers may be not the same. To maximize the reliability of stored data objects in heterogeneous networks, the optimal allocation of erasure-coded fragments is a challenging problem constrained with heterogeneous peer availabilities and redundancy overhead. This paper examines this optimal problem considered MDS erasure codes applied into distributed storage networks. First, we model the reliability of an allocation with the weighted-k-out-of-s model and extend its properties to efficiently calculate the reliability of an allocation; then we reduce the reliability computation of a given allocation to linear computation cost based on the weighted k-out-of-s model. Then, we deduce the problem to integer partition problem and propose two order-based search algorithms. Our experiments show that our proposed algorithms can be applied to find the optimal allocations efficiently in various practical coding cases. Furthermore, we evaluate the performance of our proposed search algorithms with some practical storage settings, and then present experimental results including the reliability, redundancy overheads and allocation pattern for the optimal allocation driven by practical network traces.

Peer-to-peer distributed storage systems aggregate the storage space of many peers spread over the Internet. Due to the dynamic and scalable nature of these systems, it is a challenging issue to access data in an available and reliable way through redundancy. Following the modeling methodology presented, we present the stochastic model to analyze redundancy evolution of these systems under churn. Different from the previous work based on the average peer availability, the stochastic model can be applied into the practice based on both conditional probabilities (alpha, theta) which can be obtained from network probing easily. First, we apply the model to characterize the redundancy evolution of a fragment system with temporary churn. Then, we use an empirical trace and a synthetic trace to validate the model. Second, based on the characteristics of different churn from the both probabilities, we propose the redundancy maintenance strategy assisted by network sampling. Our simulations evaluate the performance of the strategy driven by empirical and synthetic traces.

This paper analyzes churn impact on replicated data duration with different node lifetime distributions. In structured overlay networks, churn includes node-join churn and node-failure churn, caused by the arrival and departure of nodes separately. The paper introduces a duration model of replicated data under node-failure churn for node failure directly leads to data loss. Furthermore, it investigates the impact of node-join churn on the duration of replicated data for different node-lifetime distributions. The paper presents that node-churn will negatively impact on replicated data duration for heavy-tailed distribution and Weibull distribution except exponential distribution. Then we evaluate the impact on replicated data duration with two real-world trace datasets. The experimental results show the negative impact of node-join churn for different node-join churn degrees. Finally, the paper discusses an enhancement by setting a trial period for every fresh node. By experiment, it is an effective way to reduce the negative impact of node-join churn due to the memory property of node lifetime distributions.

For practical deployment of peer-to-peer (P2P) systems, it is one of the most important and challengeable aspects to achieve high data availability in structured P2P systems since the environment is much scalable and dynamic. The paper utilizes the hybrid of two data redundancy schemes, namely replication and erasure coding, to improve system availability. To mask or hide the high churn from the short-lived but churn-frequent peers and permanent failure peers, we use replication among the nodes in a certain interval of the identifier space that can be considered as a virtual node. Then with an erasure-coded redundancy scheme, we consider that a set of virtual nodes that cooperatively provide guaranteed over the networks. The paper presents the hybrid redundancy prototype and protocol. The evaluation shows that the approach is effective with an empirical trace by setting different system parameters.