SLA Requirements Research Articles

5G and edge: A reinforcement learning approach for Virtual Network Embedding with cost optimization and improved acceptance rate

5G technologies are fueling a revolution across numerous industries, including manufacturing, healthcare, and entertainment, by enabling the development and deployment of novel applications at the network’s edge. To meet the demanding service level agreements of these industries, a dynamic and adaptable infrastructure strategy that combines cloud and edge computing models is needed. This hybrid approach offers the benefits of both centralized cloud processing and decentralized edge computing, optimized for responsiveness and efficiency. A key element for success is an orchestration mechanism that dynamically allocates resources to ensure the infrastructure can adapt to fluctuating demands in real time, optimizing resource utilization and meeting SLA requirements. Among these mechanisms, virtual network embedding (VNE) and dynamic resource management (DRM) have emerged as tools for defining where and how edge technology should be used. However, current VNE approaches struggle to adapt to real-time fluctuations in demand across geographically distributed edge resources. This work introduces a novel resource allocation algorithm, the VNE-CRS, which uses an Artificial Intelligence technique called Reinforcement Learning to orchestrate resources across multiple domains. This approach benefits from the strength of Reinforcement Learning: its ability to consider the entire problem from beginning to end while incorporating various aspects of 5G Quality of Service Indicators for optimal decision-making. Experiments were conducted to evaluate the performance of VNE-CRS against state-of-the-art algorithms for multi-domain edge environments. Results have shown that employing Reinforcement Learning techniques for VNE resource allocation yields performance gains of 12.32 percentage points in comparison with the GRC algorithm and 28.80 percentage points in comparison with the base edge environment, presenting an acceptability rate closer to the Public Cloud environment with all benefits of edge environment. In conclusion, VNE-CRS offers an efficient solution for resource allocation in 5G environments, achieving superior performance and transforming the VNE architecture into a comprehensive orchestration system that optimizes infrastructure utilization for strategic long-term benefits.

Optimizing Resource Management for Shared Microservices: A Scalable System Design

A common approach to improving resource utilization in data centers is to adaptively provision resources based on the actual workload. One fundamental challenge of doing this in microservice management frameworks, however, is that different components of a service can exhibit significant differences in their impact on end-to-end performance. To make resource management more challenging, a single microservice can be shared by multiple online services that have diverse workload patterns and SLA requirements. We present an efficient resource management system, namely Erms, for guaranteeing SLAs with high probability in shared microservice environments. Erms profiles microservice latency as a piece-wise linear function of the workload, resource usage, and interference. Based on this profiling, Erms builds resource scaling models to optimally determine latency targets for microservices with complex dependencies. Erms also designs new scheduling policies at shared microservices to further enhance resource efficiency. Experiments across microservice benchmarks as well as trace-driven simulations demonstrate that Erms can reduce SLA violation probability by 5× and more importantly, lead to a reduction in resource usage by 1.6×, compared to state-of-the-art approaches.

CD-MSA: Cooperative and Deadline-Aware Scheduling for Efficient Multi-Tenancy on DNN Accelerators

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 <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Cooperative and Deadline-aware Multi-Systolic-Array scheduling</i> (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 <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">dynamic task fusion</i> , 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.

PARS: A Priority-Aware Resource Sharing Scheme for Services With Differential QoS Requirements in B5G Edge Network

Next-generation networks are expected to address scenarios demanding varying QoS, such as higher bandwidth, lower latency, and reliable data delivery. A multitude of B5G applications like AR/VR/XR and health-care needs differential QoS requirements and higher reliability of internet services. In this context, we develop a novel approach, PARS, for resource provisioning among slices for facilitating efficient resource utilization in an edge network. Further, we propose an adaptive SFC-RA algorithm to maximize user admission under computation and bandwidth resource constraints while satisfying SLA requirements of health-care services. Moreover, we show efficacy of our solution by comparing it with existing state-of-the-art methods.

Network slicing in virtualized 5G Core with VNF sharing

Through multiplexing separate virtual networks on the same network infrastructure, network slicing will lead to customization, scalability, flexibility, and isolation of services in different domains of the network. It also results in lower CAPEX/OPEX via increased utilization of network resources. In this paper, we have formulated an Integer Linear Programming (ILP) model to create network slices in the 5G virtualized core by considering the sharing of virtual network functions (VNFs) between different network slices to minimize cost. The ILP model aims to minimize the number of VNF instances and resource consumption while meeting the network slice requirements. The main motivation for sharing the 5G core VNFs among different network slices is to reduce the number of VNF instances, which results in a lower management overhead of VNFs. As solving the optimization problems is time-consuming and does not scale for larger deployments, we also developed a Heuristic Backtracking Algorithm for Network Slicing (HBA-NS). Furthermore, in order to account for the dynamic nature of traffic, our proposed network slicing framework also includes a CNN-LSTM deep learning model to predict UEs’ requests in each network slice. Based on the predicted requests, the HBA-NS algorithm scales allocated resources of different network slices to meet the dynamic SLA requirements of users. Our evaluation results show that the solution achieved by the HBA-NS algorithm is on average 6% worse than the optimal solution in the worst case but derived at a significantly lower run time. We also evaluated the proposed network slicing framework in terms of different criteria including its isolation property. Our simulation results show that network slicing can provide better performance compared to competitive approaches.

Superactive: a priority, latency, and SLA-aware resource management scheme for software defined space-air-ground integrated networks

The software-defined space-air-ground integrated network (SAGIN) is regarded as future-generation networking solution due to its wide-area coverage and seamless communication support for ground networks and resource-intensive application support for space-air networks. The SDN-based literary works are restricted only to resource management for space-air or ground networks, not both. The resource scheduling for multi-users caching, computing and data-transfer task accomplishment over software-defined SAGIN were out of their investigations by taking proactive local SDN, heterogenous users and applications, SLA requirements, priority, latency, and users' budget into account. To outperform the problems, this paper advocates a proactive SDN-based resource management scheme considering SLA requirements, latency, priority, and users' budget for multi-user task completion over software-defined SAGIN. This paper contributes an analytical model that covers task completion time, energy expenditure, financial cost, and SLA fulfilment metrics. The performance results illustrate that proposed scheme produces 72% time and 69% financial gain over the compared scheme.

A multi-objective approach for energy-efficient and reliable dynamic VM consolidation in cloud data centers

The rapid growth of cloud computing in the last decade has led to an increasing concern about the energy requirement of cloud data centers. Dynamic virtual machine (VM) consolidation is an effective way to tackle this issue, where VMs are executed on as few physical machines (PMs) as possible. Meanwhile, VM placement must be performed strategically, by considering different factors of the available resources to optimal exploitation of them. Moreover, a major challenge is the system’s reliability degradation because of the high frequency of consolidation and placing VMs on unreliable PMs. In this paper, we address the problem by introducing a discrete-time Markov chain (DTMC) model to predict future resource usage. Using the DTMC model along with the reliability model of PMs leads to more accurate PMs categorization based on their status. Then, a multi-objective VM placement approach is proposed to achieve the optimal VMs to PMs mapping using the ε-dominance-based multi-objective artificial bee colony (ε-MOABC) algorithm which can efficiently balance the overall energy consumption, resource wastage, and the system reliability to meet SLA and QoS requirements. We have validated the effectiveness of our proposed approach by conducting a performance evaluation study using the CloudSim toolkit. Competitive analysis of the experimental results demonstrates that the proposed approach significantly improves energy consumption while avoiding the inefficient VM migrations.

A green, energy, and trust-aware multi-objective cloud coalition formation approach

The workload of cloud providers is often uncertain due to the dynamic and varied client demands, thus presenting a significant challenge. If cloud providers fail to meet changing demands during peak hours, they may experience service requests’ delays and rejections, damaging both their reputation and their profit margin. The idea of cloud federations is contemplated as a solution to this challenge by allowing a group of cloud providers to combine their resources, as and when required, to satisfy client requests dynamically. Despite the merits of the existing cloud federation approaches, they suffer from several limitations, such as performance issues, the inability to converge to a solution in complex scenarios, the lack of consideration of SLA requirements during the federation formation phase, and the lack of flexibility offered to clients to prioritize specific objectives during the formation of the cloud federation. This work addresses these deficiencies by presenting a green, practical, and customizable SLA-based approach to cloud federation formation. Inspired by social gaming, the proposed approach can rapidly form a stable coalition of cloud providers able to fulfill client requests, while ensuring the satisfaction of SLA requirements in terms of performance, energy consumption, CO2 emissions, and providers’ trustfulness. At the heart of this approach lies a new coalition formation algorithm, dubbed the “multi-criterion coalition formation algorithm (MCCF).” Unlike existing collation formation algorithms, the MCCF algorithm offers clients the flexibility to prioritize the objectives used for cloud federation formation. Based on those preferences, the MCCF algorithm generates stable SLA-aware groups of cloud providers that dynamically adapt to network changes using a self-healing process. The MCCF algorithm has been implemented and extensively tested using a wide range of scenarios, and its performance was compared with that of two state of the art approaches: The optimal approach and the split-and-merge approach. The results show that the MCCF algorithm outperforms the optimal and the split-and-merge approaches in terms of coalition size, individual providers’ payoff, and mistrust cost. Moreover, the proposed approach yields a zero % request rejection rate and an execution time that is 8 to 16 times faster than the split-and-merge and optimal approaches, while enabling a reduction of the Carbon footprint of the cloud coalition operations. Owing to these advantages and its ability to offer a flexible and customizable solution to clients, the MCCF algorithm represents a promising and environmentally conscious cloud collation formation approach that is ideal for real-time applications and practical cloud deployments.

An SLA-Aware Cloud Coalition Formation Approach for Virtualized Networks

One of the main challenges faced by cloud providers is the uncertainty in their workload, resulting from the high variability and the dynamic nature of clients' demands. The inability to meet those demands during peak times can lead to high service rejection rates, experienced delays, and consequently profit and reputation losses. The concept of cloud federation has been proposed as a way to address this challenge, by enabling a group of cloud providers to collaborate by dynamically combining their resources as needed, to satisfy received requests. Existing cloud federation approaches fail to consider clients' SLA requirements during the coalition formation process or provide a self-healing mechanism to deal with unexpected resources' shortage during operation. Furthermore, the state of the art approaches suffer from performance issues, such as high execution times, unstable performance, and lack of convergence to a solution in complex scenarios (e.g., requests with mixed, independent types of VMs). This paper proposes a novel social gaming based approach for coalition formation in the cloud that finds the best coalition of cloud providers to answer requests, while satisfying the clients' SLA requirements. The proposed algorithm, dubbed SLA Aware Cloud Coalition Formation algorithm (S-ACCF), leverages Irving's roommate algorithm to form a stable coalition of cloud providers, with a rapid execution time. The S-ACCF algorithm is designed to maximize the coalition's profit, while minimizing the number of participants in the coalition as well as the penalty incurred by providers who fail to offer all or some of the promised resources using a self-healing process. The S-ACCF algorithm was extensively tested using a variety of scenarios, and its performance was compared to two state of the art approaches: 1) the Optimal Cloud Federation Mechanism (OCFM) that relies on an exhaustive search of all possible solutions to find the best coalition; and 2) the Cloud Federation Formation Mechanism (CFFM) that relies on an iterative split-and-merge approach to find the best coalition. While the optimal approach (OCFM) always finds the best coalition leading to the highest collective profit, it has an exponential time complexity, thus leading to very large execution times. On the other hand, the split-and-merge approach (CFFM), which relies on random selection of sub-groups for coalition formation, suffers from instability (different results in repeated runs), high and variable execution time, and a noticeable requests' rejection rate that changes between runs. The test results show that the S-ACCF algorithm addresses the limitations of the OCFM and the CFFM algorithms, and outperforms the optimal and split-and-merge approaches in terms of execution time, individual provider payoff, and the number of providers per coalition. Furthermore, it yields higher stability and zero rejection rate, when compared to the split-and-merge approach. Indeed, our proposed approach yields an execution time that is 12 to 25 times faster than the optimal and split-and-merge approaches, which is a major advantage for real-time applications. Moreover, when compared to the two other approaches, our S-ACCF algorithm always finds the smallest coalition possible satisfying the client requirements, thus leading to the highest individual payoff for providers and lower administration overhead. Finally, unlike the split-and-merge approach, our algorithm shows a stable performance, and converges towards the optimal solution in simple and complex scenarios, thus making it suitable for production environments.

Energy efficient VM scheduling strategies for HPC workloads in cloud data centers

In virtualized environments, virtual machine (VM) scheduling strategies incorporated with energy efficient techniques are needed to reduce the operational cost of the system while delivering high Quality of Service (QoS). It is widely accepted that the cost of the energy consumption in the environment is a dominant part of the owner’s budget. However, when considering energy efficiency, VM scheduling decisions become more constrained, leading in the violation of job deadlines and hence compromising QoS. This paper studies energy efficient VM scheduling strategies in virtualized environments to minimize the queue time and makespan under the fulfillment of SLA requirements (i.e., deadline). Specifically, six energy efficient VM scheduling strategies are investigated incorporated with the dynamic voltage and frequency scaling (DVFS) power management technique. They consist of user-oriented and system-oriented policies. The strategies are extensively simulated and compared with three power management governing methods provided at hypervisor level (i.e., userspec, ondemand, and performance). To conduct simulation experiments, we employ real-world high performance computing (HPC) workloads collected from a production data center. For comparison and evaluation, we analyze the: (a) energy consumption, (b) runtime, (c) queue time, (d) makespan, and (e) slowdown ratio. Lastly, we highlight the strengths and weaknesses of the VM scheduling strategies that can help to choose the most appropriate VM scheduling strategy for a given scenario.

Multi-prediction based scheduling for hybrid workloads in the cloud data center

Cloud computing can leverage over-provisioned resources that are wasted in traditional data centers hosting production applications by consolidating tasks with lower QoS and SLA requirements. However, the dramatic fluctuation of workloads with lower QoS and SLA requirements may impact the performance of production applications. Frequent task eviction, killing and rescheduling operations also waste CPU cycles and create overhead. This paper aims to schedule hybrid workloads in the cloud data center to reduce task failures and increase resource utilization. The multi-prediction model, including the ARMA model and the feedback based online AR model, is used to predict the current and the future resource availability. Decision to accept or reject a new task is based on the available resources and task properties. Evaluations show that the scheduler can reduce the host overload and failed tasks by nearly 70%, and increase effective resource utilization by more than 65%. The task delay performance degradation is also acceptable.

An energy efficient and SLA compliant approach for resource allocation and consolidation in cloud computing environments

Cloud computing is a promising paradigm that enables a “computing-as-a-service” model, in which a dynamic pool of virtualized computational resources (e.g. CPU) can be leased and released on demand. With the increased demand for cloud computing infrastructures and the explosion in data center sizes, energy efficiency becomes an important issue to consider. Green cloud computing is an area that focuses on the design of energy efficient data centers, in order to achieve cost savings and minimize negative impacts on the environment. One of the main green cloud computing strategies used for the reduction of energy consumption consists in maximizing the utilization of a number of physical machines (PMs) and turning off or suspending unused servers. This strategy is typically achieved using two types of algorithms: Virtual Machines (VMs) placement and VMs consolidation algorithms. VMs’ placement is a process of dynamically placing VMs onto PMs while satisfying specific VM-to-PM mapping rules. VMs’ consolidation optimizes the resource utilization and groups dispersed VMs on a minimal number of active PMs, based on live migration techniques. Both approaches are time and resource consuming, and are categorized as NP-hard optimization problems. Moreover, reducing energy consumption by means of resource consolidation may reduce the system’s availability and reliability, and lead to SLA violations. Therefore, there is a need for multi-objective optimization approaches that can strike a balance between energy consumption and the system’s ability to meet QoS and SLA requirements. In this work, we propose an energy-aware and QoS-aware multi-objective Ant Colony Optimization (MACO) approach for VM placement and consolidation. Our approach aims at achieving a trade-off between energy efficiency, system performance, and SLA-compliance. The proposed approach was implemented and tested in small to mid-size data center settings, simulated using cloudSim, and was compared with a number of heuristic and meta-heuristic approaches, using eight performance metrics. The results show that our approach outperforms the other tested approaches in terms of energy savings, reduction of resource wastage in term of CPU, reduction of communication cost in term of energy induced by traffic load exchanged between VMs, and minimization of the number of VM migrations and SLA violations – thus demonstrating an ability to balance energy efficiency with system performance and QoS requirements.

Research of Automatic Configuration Technology for Virtual Machines based on Cloud Computing

Cloud resource managers face many problems, such as the dynamic changes of incoming load and demand elasticity of resources. From the aspect of elastic configuration management technology of virtual resources, this paper focus on how to provide quick and reliable cloud resources for users. Virtual machine resources automatic configuration management technology is proposed in this paper, the reverse reinforcement learning technique is introduced into cloud virtual resource management, configuration management process of the virtual machine is modeled as a Markov decision model. According to the running state of the application system and the dynamic changes of the input load, this technology can make an automatic decision to add or remove a number of virtual machines. Experimental results show that this technology can complete the tasks of automating configuration of virtual resource management according to the changing load, respond to end user's in a timely manner, and ensure the SLA requirements of cloud users.

Multi-tenant Data Migration Strategy in SaaS Platform

On demand migration of tenants’ data is a critical technology for elastic load balancing in SaaS application. However, different tenants have different required resource with various SLA requirements. Moreover, previously reported migration techniques are not tenant aware. To study the elastic load balancing of tenants’ data, we mainly focus on the problem of when to migrate which tenants’ data in a muti-tenant shared DB server and build a new framework that aims to monitor load and minimize migration cost. We establish a demand resource estimation model to decide when to migration and then create a cost model to guide the choice of possible migration tenants. The simulation experimental results show the high effectiveness and the good efficiency of our model.

Not All Packets Are Equal, Part I: Streaming Video Coding and SLA Requirements

In this first part of a two-part article, the authors consider the network factors that impact the viewers' quality of experience (QoE) for IP-based video-streaming services such as IPTV. They describe the IP service-level requirements for a transported video service and explain MPEG encoding to help readers better understand the impact that packet loss has on viewers' QoE.

An objective-oriented service model for VoIP overlay networks over DiffServ/MPLS networks

Bandwidth provisioning and QoS mapping are key issues to support multimedia services such as VoIP in the emerging network technologies. The classification feature enables MPLS (Multi-protocol Label Switch) to support differentiated types of services (DiffServ) with needed QoS. The differentiated service model provides a variety of mechanisms to achieve different objectives (such as call level or packet level satisfaction). To design a good service model, which can balance the call level and packet level QoS performances, is a challenging task. In this paper, we propose two objective-oriented service models for the VoIP services over the DiffServ/MPLS networks. They can be modeled as the continuous time Markov chains (CTMC) and the performance are assessed in details. The salient point for the proposed service models is to solve the myth of the trade-off between the service quality (user’s concern) and the system revenue (system provider’s concern) – involving how to meet each user’s SLA requirements while maximizing system revenue. The analytical results in this paper can provide useful information to both user and service provider for signing a cost-effective contract which provides a better trade-off between cost and QoS.