YHYi Hu

Yi Hu

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Yi Hu has a total of 48 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Microservice Autoscaling With Fine-Grained DVFS in Dynamic Multi-Edge Collaboration via Deep Reinforcement Learning

    by Bo Xu, Ruyao Zou, Jiangtian Nie, Zhiheng Yao, Yi Hu, Hao Wen, Menglan Hu, Kai Peng

    Microservice architecture has become the mainstream approach for low-latency and robust edge scenarios. In practice, service demands exhibit high temporal variability and concurrency, making efficient autoscaling significantly critical. However, existing research on autoscaling adopts coarse-grained service models that treat instance adjustment and request routing in isolation. This simplification neglects the intricate interdependencies among services, thus undermining the efficacy of autoscaling. In addition, many current autoscaling models do not simultaneously account for autoscaling and dynamic voltage and frequency scaling (DVFS), making it difficult to achieve an effective balance between performance and energy consumption. To overcome these challenges, this paper proposes an autoscaling algorithm that synergistically combines deep reinforcement learning with DVFS. First, we build a multi-instance performance model grounded in queuing theory, formulating microservice autoscaling as an optimization that minimizes both latency and energy consumption. Second, we present the Reinforcement Learning and DVFS Joint Optimization Autoscaling Algorithm (RLDA), which observes system state changes and scales microservice instances in real time while jointly optimizing DVFS. Third, to balance latency and energy consumption, we design the Latency-Constrained Greedy Frequency Scaling Algorithm (LGFS), which adjusts server operating voltage based on deployment scheme. Extensive simulation results demonstrate that under the Twitter traffic pattern, our algorithm achieves comprehensive performance improvements in energy consumption and latency of 37%, 25%, and 29% compared to DRL, ProScale, and PPA, respectively. Under the World Cup traffic pattern, the gains are 33%, 28%, and 23%, respectively.

  • Microservice Orchestration With Database for Delay-Sensitive Applications in Cloud–Edge Networks via Deep Reinforcement Learning

    by Yi Hu, Hangyu Cheng, Hao Li, Fuwei Guo, Hao Wang, Menglan Hu, Kai Peng

    Emerging microservice architectures have enabled the efficient development and maintenance of a wide variety of online applications. However, the computationally intensive nature of these applications poses substantial challenges for microservice orchestration in cloud-edge collaborative networks. The frequent communications and complex service dependencies between microservices create tight coupling between deployment and routing, complicating the orchestration process. Furthermore, large-scale stateful microservices often require numerous data queries to cloud-based databases, resulting in significant communication overhead, which adversely affects quality of service (QoS). While many existing studies have focused on isolated aspects of this problem, such as reinforcement learning based orchestration and edge-only scheduling frameworks, a comprehensive solution addressing both computational and database access challenges is still lacking. To bridge this gap, we investigate microservice orchestration in hybrid edge-cloud networks with a focus on database interactions for delay-sensitive applications. We propose a tree-based application model and multi-instance queuing networks for precise delay analysis. To optimize performance, we introduce the Tree Structure Chain Serialization (TCSVN) algorithm, integrated with reinforcement learning using truncated Gaussian Thompson sampling and a step-updating mechanism. Finally, experimental results demonstrate that our approach improves end-to-end latency by 32%-45% and reduces deployment cost by 14%-26% compared to existing baseline methods across multiple performance metrics.

  • Hybrid Orchestration of AI Services and Microservices in Multi-Edge Collaboration Based on Lagrangian Relaxation Deep Reinforcement Learning

    by Jiaxiang Xu, Zhaoyi Wang, Di Han, Yi Hu, Fuming Fan, Menglan Hu, Kai Peng

    The rapid development of artificial intelligence has promoted the application of AI services in various fields. The combination of the microservice architecture and the “AI as a Service” (AaaS) to form an intelligent microservice system can realize the orchestration of intelligent applications. In intelligent applications, AI services often require the key functions of accompanying microservices, such as service discovery, data collection, gateway routing, etc., to ensure the stability and efficiency. Otherwise, it will lead to the incompleteness of applications, severely affecting the overall performance of the system. Therefore, fine-grained hybrid orchestration of AI services and microservices is crucial. Nevertheless, it faces major challenges in the resource-constrained edge environment: complex dependencies, frequent data communication, and heterogeneous resource competition between microservices and AI services. Moreover, the tight coupling between deployment and routing leads to complex joint optimization problems. Considering either deployment or routing separately will lead to poor system performance. However, most of the existing studies ignore the above problems.To this end, this paper investigates the hybrid orchestration problem. First, we model intelligent applications as hybrid service invocation chains. Then, based on Jackson queuing networks and multi-instance models, we deeply study the complex dependencies among hybrid services and the heterogeneous resource allocation problem, and conduct an accurate analysis of the request delay and system energy consumption. Furthermore, we innovatively propose Lagrangian Relaxed Deep Deterministic Policy Gradient (LR_DDPG), which introduces a cost mechanism and transforms the constrained hybrid orchestration process into an unconstrained optimization problem by Lagrange relaxation to find the optimal deployment and routing strategy under complex constraints. Finally, the experimental results show that our algorithm achieves significant advantages in terms of latency, energy consumption, and request success rate.


  • Collaborative Orchestration of Microservices and AI Services in Edges: A Dual-Time-Scale Reinforcement Learning Approach

    by Kai Peng, Junhui Hu, Zhiyu Wang, Yi Hu, Menglan Hu, Zehui Xiong, Zhe Chen

    The rapid development of service computing has led to the emergence of scalable and flexible architectures such as microservices and Artificial Intelligence as a Service (AaaS), enabling the orchestration of AI-driven intelligent applications. However, existing work on intelligent applications orchestration overlooked essential microservice components that support AI services, resulting in coarse-grained and incomplete models. To ensure system integrity and enhance QoS, fine-grained collaborative orchestration of microservices and AI services is crucial. However, this poses significant challenges due to complex service dependencies, high request concurrency, and heterogeneous resource demands in edge environments. Moreover, the strong coupling between service deployment and request routing complicates their joint optimization, since effective decisions in one depend on the other. To address these challenges, we propose a collaborative orchestration framework that jointly optimizes the deployment of microservices and AI services along with probabilistic request routing in edge environments. We formulate the problem as a mixed-integer nonlinear program and leverage Jackson queuing networks for accurate delay modeling. To solve this, we develop a dual time scale hybrid greedy proximal policy optimization (DTS-HGPPO) algorithm that performs instance-level deployment and adaptive routing, enhanced with iterative instance planning, action masking and intrinsic motivation mechanisms. Extensive trace-driven experiments demonstrate that our method significantly reduces both response delay and service cost compared to state-of-the-art baselines.


  • Top Cited Papers

    Energy-Aware Service Mesh Deployment and Online Request Routing in Edge: A Hierarchical Deep Reinforcement Learning Approach

    by Yi Hu, Junhui Hu, Yifei Yang, Menglan Hu, Kai Peng, Tianyue Zheng, Chao Cai, Zehui Xiong

    Service meshes built upon ubiquitous microservice architectures, as an emerging paradigm, promise to enhance the flexibility, scalability, and portability of energy-consuming and latency-sensitive applications in edge with limited resources. However, due to intricate microservice dependencies, service multiplexing, and parallel distributed instances, microservice deployment and request routing are highly interdependent. To reduce response latency and energy consumption, such collaborative optimization for efficient service mesh orchestration is necessary, but significantly challenging. Besides, strict service level objective (SLO) requirements and f ine-grained latency analysis with multi-nest routing further impose great difficulties to online orchestration. When considering multi instance modeling and multi-hop data communications for numerous microservices, the difficulty is extremely amplified. Nevertheless, most prevailing work failed to design sophisticate models and methods for addressing the above difficulties, and ignored the inherent transmission energy consumption for highly-concurrent multi-hop data interactions. Therefore, this paper investigates the energy aware service mesh deployment and online request routing in edge. First, we establish a multi-instance queuing network model to accurately analyze the end-to-end response latency with complicated dependencies and multi-hop communications, and optimize energy consumption in a fine-grained manner. Then, to boost the overall performance, we design an efficient multi-dimensional hierarchical deep reinforcement learning algorithm, which enables edges and service instances to cooperate with each other to handle massively concurrent requests. Besides, we propose an energy-aware proactive autoscaling algorithm to carefully adapt to exceedingly dynamic scenarios. Finally, extensive experiments are performed to show our superior performance compared to other baselines.


  • Joint Deployment and Routing for Hybrid AI Services and Microservices in Edge via Deep Reinforcement Learning

    by Shudong Zhang, Yifei Yang, Fuwei Guo, Yi Hu, Menglan Hu, Kai Peng, Chao Cai, Zehui Xiong

    The big data era has accelerated the development of artificial intelligence (AI). The model-as-a-service (MaaS) paradigm has been used to address the substantial challenges associated with the organization and development of AI services. However, the successful delivery of complete AI applications is contingent upon the robust collaboration between microservice architectures and AI services. In this case, hybrid orchestration of AI services and microservices is highly necessary, but it still brings challenges. Furthermore, due to the heterogeneity of servers, resource competition, and multi-instance, the difficulty of hybrid orchestration modeling is increased. When considering intricate service dependencies among AI services and microservices, the tight coupling of deployment and routing leads to complex joint optimization problems, vastly aggravating the pressure of hybrid orchestration. Nonetheless, extant literature largely failed to address the intricate competitive and collaborative relationships between AI services and microservices, and fine-grained latency analysis with multi-instance modeling in the hybrid orchestration problem. Therefore, we study joint deployment and routing for hybrid AI services and microservices in heterogeneous edge. First, we conduct a precise analysis of latency and energy consumption, based on queuing networks and multi-instance models. Second, we propose a reinforcement learning (RL) method based on potential functions and segmented rewards (PS_SAC) to optimize end-to-end latency and system energy consumption, achieving efficient hybrid orchestration. Finally, through extensive simulation experiments, the algorithm demonstrates significant advantages in reducing latency, improving resource utilization, and lowering system energy consumption.


  • Energy-Efficient Microservice Orchestration for Latency-Sensitive Applications in Edge Computing: A Reinforcement Learning Approach

    by Yi Hu, Xi Liao, Junhui Hu, Hao Wen, Huan Xu, Kai Peng, Menglan Hu

    With the rapid development of the Internet of Things (IoT), emerging microservice architectures support scalable and flexible deployment to handle a significant increase in latency-sensitive requests in resource-limited edge. Due to complex data dependencies and instance sharing among microservices, service deployment and request routing tightly coupled, which motivates a complex joint optimization problem. In this case, the difficulty of service orchestration is extremely enlarged when considering multi-instance modeling and fine-grained latency analysis for the strict quality of service (QoS). Furthermore, since the objectives of end-to-end response latency and energy consumption present an inherent conflicting relationship, carefully reconciling the trade-offs between them through efficient microservice orchestration is necessary, but significantly challenging. However, most previous work failed to propose suitable approaches and models to address the above difficulties. Therefore, this paper investigates the energy-efficient microservice orchestration for latency-sensitive applications in edge computing, aiming to minimize both latency and energy consumption. We first develop a multi-instance queuing network model that accurately captures data dependencies and enables detailed analysis of queuing, computing, and communication delays. Then, we propose the Energy-efficient Determination of Microservice Instances (EDMI) algorithm to derive the minimal required number of instances. Furthermore, we design a deep reinforcement learning-based orchestration method, Global-Reward Soft Actor-Critic (GRSAC), which integrates local reward signals into global feedback to mitigate sparse rewards. Finally, we provide theoretical analysis on time complexity and convergence. Experimental results show that our proposed algorithm significantly outperforms existing approaches in reducing both response latency and energy consumption.


  • Symmetric Orchestration Under Service Mesh Paradigm: Empowering Massive Online Applications in Edge Clouds

    by Kai Peng, Tongxin Liao, Mingyuan Ren, Yi Hu, Liangliang Wu, Menglan Hu, Hongbo Jiang

    With the rapid advancement of edge computing, service mesh has emerged as a critical technology for improving network performance, owing to its flexibility and scalability. However, massive online applications in edge clouds pose significant challenges to microservice orchestration, including high concurrency, complex service dependencies, strict response delay requirements, and fast orchestration needs. Addressing these challenges requires efficient and fast orchestration strategies, but existing approaches often lack accurate models and effective algorithms to handle these complexities. To tackle the above challenges, this paper proposes an efficient Symmetric Microservice Deployment (SMD) algorithm for fast orchestration. First, accurate modeling is achieved with the queuing network, which analyzes intertwined requests and calculates detailed delays. Moreover, the SMD algorithm simplifies the coupling between deployment and routing by considering internal dependencies during deployment. This integrated approach eliminates the need for separate routing solutions and ensures provable optimal performance under symmetric deployment. Experimental results demonstrate that, compared to four baseline algorithms, the proposed method reduces response delay by 25.5% and execution time by 58.4%, showcasing the potential and advantages of the algorithm for optimizing microservice orchestration in edge clouds networks.


  • Time-Varying Microservice Orchestration With Routing for Dynamic Call Graphs via Multi-Scale Deep Reinforcement Learning

    by Yi Hu, Liangbo Hou, Junhui Hu, Mingyuan Ren, Menglan Hu, Chao Cai, Kai Peng

    Lightweight microservices as a software architecture have been widely adopted in online application development. However, in highly-concurrent microservice scenarios, frequent data communications, complex call dependencies, and dynamic delay requirements bring great challenges to efficient microservice orchestration. In this case, service deployment and request routing are interactively-coupled in multi-instance modeling, and cannot be locally optimized effectively, thereby enlarging the difficulty for collaborative orchestration. To accommodate time-varying request properties and dynamic microservice multiplexing, orchestration schemes are frequently adapted to real-time parallel request queues, further complicating the difficulty. Nevertheless, most previous work failed to propose appropriate models and methods for the above issues. Therefore, this paper investigates the online microservice orchestration with probabilistic routing for dynamic call graphs in clouds. First, we formulate the time-slot-based joint optimization problem as a Markov Decision Process. The open Jackson queuing networks are used to accurately establish multi-instance models and analyze the request queuing, processing, and communicating delays. Then, we propose an efficient curiosity-driven deep reinforcement learning algorithm, which meticulously implements instance-level orchestration through multi-dimensional collaborative decisions and multi-timescale trigger events. Finally, through comprehensive trace-driven experiments, our proposed approach significantly outperforms other baselines in terms of orchestration cost and resource utilization.