Data centers rely on Priority-based Flow Control (PFC) to achieve lossless data transmission in Ethernet networks. To avoid buffer overflow, PFC pauses flows in a coarse-grained manner, which brings potential problems, e.g., Head-of-Line (HoL) blocking, and PFC deadlock. Although the state-of-the-art approach BFC with per-flow backpressure tackles some of the limitations of PFC, it faces implementation challenges due to the need for a large number of queues. In this paper, we present F-PFC, a fine-grained flow control scheme that only leverages a small amount of queues to address the limitations of PFC. Specifically, F-PFC first designs a fine-grained flow backpressure scheme to adjust the intensity of flow control adaptively. With different levels of flow backpressure, F-PFC ensures high throughput and low latency simultaneously. Then, F-PFC presents an accurate flow identification scheme to locate flows that really contribute to congestion. Finally, F-PFC presents a dynamic queue assignment and scheduling scheme to isolate congestion flows with limited queues. We theoretically analyze the performance of F-PFC and present the implementation of F-PFC. Extensive testbed experiments and large-scale simulations verify the performance of F-PFC. The experimental results show that F-PFC reduces tail latency by at least 33% and queue occupancy by 46% compared with state-of-the-art approaches.

Bare-metal cloud services provide direct access to dedicated physical hardware, significantly enhancing computational power, disk I/O, and network I/O performance. To effectively manage physical resources, bare-metal typically relies on specialized cloud management platforms. However, the existing management architecture still faces significant bottlenecks. These bottlenecks include slow and cumbersome deployment processes, inadequate security isolation that exposes the system to potential vulnerabilities, and limited scalability that fails to meet dynamic and evolving demand. Therefore, optimizing the management architecture to improve deployment efficiency, security, and flexibility has become a key challenge for bare-metal cloud services. This paper proposesNBBM(NebulaMatrix Bare Metal), an innovative bare-metal cloud management platform architecture designed to restructure the management of bare-metal servers in OpenStack. To simplify the complexity of bare-metal cloud management and significantly improve the overall system efficiency,NBBMadopts the following technologies: an architecture that thoroughly decouples compute and storage, a distributed management system based on SmartNIC technology, and a high-performance cloud storage interconnect solution relying on SmartNICs. These technological innovations enable theNBBMarchitecture to provide a more secure and efficient cloud service management solution. Extensive experimental results demonstrate that theNBBMplatform achieves minute-level deployment and delivers at least a 12× speedup (approximately 92% reduction) over widely used methods, while ensuring secure and flexible access to storage resources without compromising performance.

Edge caching is a critical application scenario in edge networks. By storing diverse files on edge servers and dynamically fetching new files from the cloud, edge networks can provide low-latency file access services for mobile users. In practice, the file fetching latency is non-negligible. Consecutive requests for the same missing file during the fetching phase introduce additional latency (referred to as delayed hits). Existing studies either ignore the delayed hits when making caching decisions or are not applicable to multi-server edge networks. In this paper, we investigate the online caching problem with delayed hits in the multi-server edge networks and prove its hardness. The objective is to minimize the total file access latency. To solve the proposed problem, we propose Cadle, which makes caching decisions based on the latency of different file access operations and weights of files in an online manner, without relying on any prior knowledge of future requests. We prove the competitive ratio of Cadle.We also conduct extensive experiments on the real-world dataset to verify the performance of Cadle. The experimental results show that Cadle reduces the total file access latency by at least 31.8% on average, and improves the hit ratio by at least 25.9% on average compared with state-of-the-art approaches.

As distributed applications increasingly span geographically distributed data centers, the demand for high-performance, long-distance transmission has been continuously growing. While intra-data-center networks have employed techniques like remote direct memory access (RDMA) to meet these design goals, extending these techniques toWANs presents unique challenges. WANs notably suffer from inherent packet losses due to buffer overflows in routers and switches, leading to decreased throughput and making distributed applications barely usable. This paper proposes Lossless Transmission (LLT), a novel buffer management scheme for enabling lossless WAN transport. LLT intelligently integrates on-chip switch buffers with an off-chip caching system to absorb traffic bursts that would otherwise cause packet loss. Its data plane logic uses a multi-level threshold system to selectively offload only critical flows during congestion. A closed-loop control protocol, managed by a stateful flow table, ensures these offloaded packets are later re-injected with guaranteed lossless and in-order delivery, effectively protecting latency-sensitive applications from retransmission overhead. We evaluate LLT using both ns-3 simulations and P4-programmable devices. The experimental results show that in typical use cases (RTT > 30ms), LLT improves link bandwidth utilization by 1.9% to 29.5% and reduces the P99 percentile tail latency by 17% to 66% in WANs compared to the state-of-the-art solutions. Overall, LLT provides a scalable, efficient, and reliable framework for long-distance data transmission, addressing critical challenges in WANs. Additionally, LLT eliminates the need for expensive WAN infrastructure modifications.

The dramatic rise in encrypted traffic brings huge challenges to traditional traffic classification methods. Deep learning-based traffic classification methods have been demonstrated to significantly improve performance. However, the following limitations remain: i) It is challenging to concurrently focus on both global and local information in traffic flows, resulting in the absence of important information. ii) The existing methods relying on temporal information suffer from low robustness in case of packet disordering or loss. iii) The use of multi-layer encryption and random routing in Tor technology poses more challenges for traffic identification. In this paper, we propose a novel ViT-based model for more accurate encrypted traffic classification, called SRViT to overcome the above challenges. Firstly, SRViT proposes a novel mechanism of multi-size patch division to learn comprehensive hidden knowledge and dependencies between packets. Secondly, we propose a self-attention operation with a relative position bias to learn the relative position relationship. After that, an incremental update mechanism is proposed to adapt to dynamic changes in the real traffic environment. At last, the comprehensive experiments on 5 real-world encrypted traffic datasets are carried out. The experimental results indicate that SRViT outperforms the state-of-the-art methods with an average accuracy improvement of 24.62% while keeping higher robustness and execution efficiency.

In recent years, large language models (LLMs) have shown great potential in many fields. LLMs deployed in cloud data centers are increasingly unable to meet the low-latency inference requirements of massive mobile users. Benefiting from various LLM lightweighting techniques and the continuously improving performance of edge servers, deploying LLMs on edge servers closer to mobile users and executing inference tasks locally can effectively reduce inference latency. However, edge servers have limited storage capacity, and deploying LLMs on edge servers incurs additional deployment overhead. In this paper, we propose an efficient LLM edge collaboration deployment strategy called EdgeColl, aiming to jointly optimize inference latency and LLM deployment costs. Specifically, EdgeColl adopts Low-Rank Adaptation (LoRA) to divide each LLM into a base model and a LoRA matrix. We formulate the LLM edge collaboration deployment problem with LoRA. Then, we present the base model deployment (BMD) strategy to achieve low inference latency and deployment costs. The LoRA deployment (LMD) strategy is also proposed to enable personalized inference. We evaluate the performance of EdgeColl. The experimental results show that EdgeColl effectively reduces LLM inference latency and deployment costs.

The state-of-the-art storage structure relies on the NVMe devices and SmartNICs to provide high IO performance and low CPU overhead. In data centers, the existing data transmission control and storage methods are not ideal, resulting in long flow completion time, especially for small IO, which directly affects the performance of disaggregated storage systems. In this paper, we present Thunder, a disaggregated storage solution designed to minimize tail latency. Firstly, Thunder achieves the minimum I/O tail latency for disaggregated storage via packet-level write-through, and has an ingenious mechanism for precise semantic conversion from message level to packet level. It refers to the process of converting message level data into packet level data and ensuring the integrity and reliability of data transmission. This process involves steps such as message segmentation, addressing, acknowledgment, and reassembly. Secondly, we present a novel optimization approach for end-to-end and information transmission processes, aiming to address a range of issues such as user usage, congestion control, and system compatibility. Finally, we conducted both testbed and large-scale simulations to verify the performance of Thunder. The results show that Thunder reduced the average latency and tail latency by 71.6% and 59.7%, respectively compared to Gimbal and Timely. Furthermore, it effectively avoids queue head blocking and congestion diffusion in PFC, increasing throughput by 2.5X and reducing tail latency by an average of 49.7%.

As a centralized location for computer systems, data centers provide high-performance computing hardware, storage devices, and network facilities for collaborative computing. The node-disjoint paths can be used to implement multi-path transmission in data center networks, which provide multiple high-quality transmission paths and improve the performance of the network. Moreover, the disjoint paths can also provide redundant transmission paths, which enhance the fault tolerance of the network. The EHDC network is a novel server-centric and highly scalable data center network based on exchanged hypercube, and its logical structure is ED(s, t). In this paper, we propose the algorithm NDPath to construct the node-disjoint paths between two distinct nodes when the two nodes are in the same EDs in ED(s, t). Moreover, we analyze the maximum length of the disjoint paths. Experimental results show that our proposed algorithm performs better than the classical algorithm Dijkstra in the Average Running Time (ART) and is very close to that in the Average Path Length (APL).

In the era of information explosion, the scale of data center networks (DCNs) has expanded exponentially, consequently leading to an inevitable increase in server failures. Therefore, how to ensure the efficient and secure operation of the network has emerged as a critically important research topic. WaveCube is a scalable, fault-tolerant, high-performance optical DCN architecture. In this paper, we first propose a local secure model (LS model) of WaveCube. This model segments fault-free nodes within sub-Wavecube by imposing specific constraints, thereby adeptly circumventing potential communication impediments that could arise due to faulty nodes. Secondly, based on this model, we design a protection routing with secure mechanism to ensure stable communication within WaveCube. Finally, we perform a series of experiments, and the results show that when the number of faulty nodes is less than half of the number of total nodes, the hit rate can reach nearly 100%, while the shortest path rate can achieve up to 90%.

Image embedding, being a fundamental task in computer vision, plays a crucial role in various downstream tasks such as image retrieval. Widely adopted in e-commerce and social media collaboration, image retrieval benefits greatly from representations learned by the embedding model. However, conventional embedding models are often trained on a single domain, leading to inadequate performance in the multi-domain scenario. To address this challenge, we introduce a generalized image embedding model designed for multi-domain image retrieval. The proposed method employs a contrastively learned Vision Transformer and a carefully crafted training scheme to enhance domain generalization capability. Our theoretical analysis and experimental results, conducted on a large-scale, real-world multi-domain image retrieval dataset, demonstrate the superiority of the proposed method over existing embedding models in terms of both accuracy and domain generalization capability.

Edge computing (EC) provides low-latency services by deploying edge servers close to users. However, these servers are prone to failures that can invalidate any predefined user allocation strategies. To ensure continuous services and maintain users' payments, affected users who are disconnected from the failed edge servers need to be reallocated. Unfortunately, due to the strict latency requirements of users and the limited resources on edge servers, many of them fail to be reallocated. Thus, we propose to migrate unaffected users from affected users' nearby edge servers to free up more resources for reallocation. In this paper, with the aim of maximizing the overall revenue and ensuring continuous service provisioning for users, we formulate the problem of Migration-Assisted _User _Reallocation (MUR) upon edge server failures and prove its NP-hardness. We then introduce an Integer Programming-based approach named MUR-O to find the optimal solution and a heuristic approach named MUR-H to efficiently find sub-optimal solutions. Experimental results on real-world datasets demonstrate that our approaches are superior to three representative approaches.

Network virtualization (NV) has extensive and significant applications in cloud computing and parallel and distributed systems. Virtual network embedding (VNE) is a key issue in NV, which is an effective means to advance systems’ performance. While existing VNE research lacks resource allocation coordination between mappings of different virtual network requests, resulting in insufficient resource utilization and high overhead. In this article, we propose a novel node essentiality evaluation model for data center networks (DCNs), and design an efficient distributed collaborative virtual network embedding. Firstly, we propose a node essentiality evaluation scheme based on dynamic model, which combines the characteristics of network topology and nodes to make the evaluation results more comprehensive. Secondly, we establish the two-stage node importance evaluation criteria for the deviation mean of the data center dynamic model and the variance based on the deviation mean. Furthermore, we investigate a nodal importance assessment method based on the data center dynamic model for perturbation testing. Finally, we design a distributed coordinated VNE algorithm (CNI-VNE) which calculates the importance index of physical nodes through topology awareness. The proposed algorithm can increase the coordination between different request mappings, thereby reducing the mapping cost of physical node resources and minimizing the cost of VNE. We use the real Fat-tree DCN of 128 servers and 80 switches as testbed, and evaluate them from indicators such as average reliability, average bandwidth consumption, average energy consumption, and average mapping time. Massive simulation results in different scenarios show that our algorithm achieves the best performance on most indicators compared with the existing state-of-the-art proposals, mapping acceptance and average revenue increased by 19.4% and 21.3%, respectively, and DCN reduced bandwidth consumption by about 30%.

Mobile edge computing (MEC) is emerging as a novel computing paradigm that pushes network resources (such as computation and storage resources) away from the centralized data center to distributed edge servers. By hiring various resources of nearby edge servers, the MEC provides high-bandwidth and low-latency network services for mobile users. As numerous mobile users may compete for limited edge servers’ resources, to improve the resource utilization of the MEC system, it is very critical to investigate an effective user allocation policy. Previous studies mainly focus on investigating offline user allocation policies. However, mobile users may arrive online, and the MEC should be able to allocate these users online too. In a real-world MEC environment, online allocation decisions should not be made entirely in the dark. The historical user requests which may contain powerful hints about future user requests, can be adopted to assist in making allocation decisions. In this paper, we take the historical data into account and study the history-assisted online user allocation strategy. Specifically, we formulate the user allocation problem with a comprehensive model and show its hardness. Then, we present an online algorithm named HOUA to allocate mobile users according to both the online arrived user requests and the historical user requests. The competitive ratio of HOUA is proved. To further verify the effectiveness of HOUA, we conduct experiments on a widely-used real-world dataset. We show that HOUA can allocate more mobile users and achieve high resource rental revenue compared with the other approaches.

Network update enables Software-Defined Networks (SDNs) to optimize the data plane performance. The single update focuses on processing one update event at a time,i.e., updating a set of flows from their initial routes to target routes, but it fails to handle continuously arriving update events in time incurred by high-frequency network changes. On the contrary, the continuous update proposed in “Update Algebra” can handle multiple update events concurrently and respond to the network condition changes at all times. However, “Update Algebra” only guarantees the blackhole-free and loop-free update. The congestion-free property cannot be respected. In this paper, we propose Coeus to achieve the continuous update while maintaining consistency,i.e., ensuring the blackhole-free, loop-free, and congestion-free properties simultaneously. Firstly, we establish the continuous update model based on the update operations in update events. With the update model, we dynamically reconstruct the operation dependency graph (ODG) to capture the relationship between update operations and link utilization variations. Then, we develop a composition algorithm to eliminate redundant operations in update events. To further speed up the update procedure, we present a partition algorithm to split the operation nodes of the ODG into a series of suboperation nodes that can be executed independently. The partition algorithm is proven to be optimal. Finally, extensive evaluations show that Coeus can improve the update speed by at least 179% and reduce redundant operations by at least 52% compared with state-of-the-art approaches when the arrival rate of update events equals three times per second.

Edge computing provides task offloading services to extend the computational capacity of mobile users and reduce task latency. The deadline-awareness offloading algorithm plays a key role in guaranteeing the quality of service (QoS) requirement. Prior studies mainly focus on tasks with strict deadlines. However, some tasks may not always have to be finished before hard deadlines, e.g., multimedia tasks. Tasks with soft deadlines can miss their primary deadlines, but not by too much. This has not been properly considered by existing offloading approaches. In this paper, we propose CONFECT to offload tasks with mixed deadlines. We formulate the problem and prove its hardness. Then, we propose two online algorithms with proven competitive ratios to solve the problem collectively, including an algorithm that assigns tasks to edge servers and an algorithm that adjusts the task execution order on each server. Extensive experiments show that CONFECT outperforms five baseline algorithms.

Although the logically-centralized perspective is offered in Software-Defined Networking (SDN), the data plane is still distributed. Update commands sent by the centralized controller are executed asynchronously and independently in each switch. The timed SDN enables synchronous and coordinated update operations as each update command can be triggered by a pre-defined timestamp. Prior work on timed update mainly focuses on producing a congestion-free update sequence, whereas finding a congestion-free timed update sequence may prolong the update time. Even worse, such an update order may not exist. In this paper, we propose Chronus+, a novel timed update system that utilizes the switch buffer to shorten the update time while minimizing the switch buffer during updates. First, we formulate the Minimum Switch Buffer Problem (MSBP) as an integer linear programming and show its hardness. Then, we propose a set of efficient algorithms to solve the problem in polynomial time. Extensive evaluations in Mininet and large-scale simulations show that Chronus+can reduce the update time by at least 17% and the switch buffer size by at least 27% compared with state-of-the-art approaches.

Edge computing is an emerging technology that provides the ability to offload tasks from resource-limited mobile devices to edge servers. Offloading tasks effectively can reduce service latencies of tasks and save resources of mobile devices significantly. Existing researches for task offloading mainly focus on how to offload tasks produced by latency-sensitive applications to meet their deadline requirements strictly. However, the research for offloading tasks produced by computation-intensive applications which do not require strict completion time is ignored. In this paper, we take latency-sensitive applications (i.e., tasks with hard deadlines) and computation-intensive applications (i.e., tasks with soft deadlines) into consideration simultaneously and design an offloading strategy to maximize the number of tasks that meet deadlines. We use a motivating example to illustrate the proposed problem. Then we introduce the solution of the problem briefly.

Although the logically-centralized perspective is offered in Software-Defined Networks (SDNs), the data plane is still distributed in nature. Update commands sent by the centralized controller are executed asynchronously and independently in each switch. The timed SDNs enable synchronous and coordinate update operations as each update command can be triggered by a pre-defined time point. Prior work on timed update mainly focuses on how to produce a congestion-free update sequence, whereas finding a congestion-free timed update sequence may be too long to be applied in practice, even worse, such an update order may not exist. In this paper, we propose Chronus+, a timed update system that utilizes switch buffer to shorten the update time while minimizing the switch buffer size during updates. We formulate the Minimum Switch Buffer Size Problem (MSBSP) as an optimization program and show its hardness. A set of efficient algorithms is proposed to determine a timed update sequence in polynomial time. Extensive evaluations in Mininet and large-scale simulations show that Chronus+can reduce the update time by at least 17% and the switch buffer by at least 27% compared with state-of-the-art approaches.

Network update enables Software-Defined Networks (SDNs) to optimize the data plane performance via southbound APIs. The single update between the initial and the final network states fail to handle high-frequency changes or the burst event during the update procedure in time, leading to prolonged update time and inefficiency. On the contrary, the continuous update can respond to the network condition changes at all times. However, existing work, especially "Update Algebra" can only guarantee blackhole- and loop-free. The congestion-free property cannot be respected during the update procedure. In this paper, we propose Coeus, a continuous network update system while maintaining blackhole-, loop- and congestion-free simultaneously. Firstly, we establish an operation-based continuous update model. Based on this model, we dynamically reconstruct an operation dependency graph to capture unexecuted update operations and the link utilization variations. Subsequently, we develop an operation composition algorithm to eliminate redundant update commands and an operation node partition algorithm to speed up the update procedure. We prove that the partition algorithm is optimal and can guarantee the consistency. Finally, extensive evaluations show that Coeus can improve the makespan by at least 179% compared with state-of-the-art approaches when the arrival rate of update events equals to three times per second.