Virtual CPU Research Articles

Exploring Performance Degradation in Virtual Machines Sharing a Cloud Server

Cloud computing has become a leading technology for IT infrastructure, with many companies migrating their services to cloud servers in recent years. As cloud services continue to expand, the issue of cloud monitoring has become increasingly important. One important metric to monitor is CPU steal time, which measures the amount of time a virtual CPU waits for the actual CPU. In this study, we focus on the impact of CPU steal time on virtual machine performance and the potential problems that can arise. We implement our work using an OpenStack-based cloud environment and investigate intrusive and non-intrusive monitoring methods. Our analysis provides insights into the importance of CPU steal time monitoring and its impact on cloud performance.

V-Recover: Virtual Machine Recovery When Live Migration Fails

Live migration is a critical technology used in cloud infrastructures to transfer running virtual machines (VMs). When live migration fails, as it often does, it is critical that any VMs in transit are not lost. There are two primary live migration techniques – pre-copy and post-copy. Pre-copy transfers a VM's memory to the destination before its virtual CPUs are transferred, whereas post-copy does the reverse. Both pre-copy and post-copy will lose the VM if the source machine fails during migration. Additionally, post-copy can lose the VM if the destination machine or network fail since the VM's memory and execution state are split across the source and destination machines. We present V-Recover, an approach to recover a VM when the source, destination, or network fails during live migration. V-Recover consists of two techniques: (1) a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">forward incremental checkpointing</i> (FIC) mechanism to handle source machine failure during both pre-copy and post-copy, and (2) a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">reverse incremental checkpointing</i> (RIC) mechanism to handle destination or network failure during post-copy. We present the design, implementation, and evaluation of V-Recover in the KVM/QEMU virtualization platform. Our evaluations show that V-Recover effectively recovers a VM upon migration failure with acceptable overheads on migration metrics and application performance.

Optimal resource optimisation based on multi‐layer monitoring

AbstractThe satisfaction of the Quality of Service (QoS) levels during an entire service life‐cycle is one of the key targets for Service Providers (SP). To achieve this in an optimal way, it is required to predict the exact amount of the needed physical and virtual resources, for example, CPU and memory usage, for any possible combination of parameters that affect the system workload, such as number of users, duration of each request, etc. To solve this problem, the authors introduce a novel architecture and its open‐source implementation that a) monitors and collects data from heterogeneous resources, b) uses them to train machine learning models and c) tailors them to each particular service type. The candidate solution is validated in two real‐life services showing very good accuracy in predicting the required resources for a large number of operational configurations where a data augmentation method is also applied to further decrease the estimation error up to 32%.

CfSNV: a software tool for the sensitive detection of somatic mutations from cell-free DNA.

Cell-free DNA (cfDNA) in blood, viewed as a surrogate for tumor biopsy, has many clinical applications, including diagnosing cancer, guiding cancer treatment and monitoring treatment response. All these applications depend on an indispensable, yet underdeveloped task: detecting somatic mutations from cfDNA. The task is challenging because of the low tumor fraction in cfDNA. Recently, we developed the computational method cfSNV, the first method that comprehensively considers the properties of cfDNA for the sensitive detection of mutations from cfDNA. cfSNV vastly outperformed the conventional methods that were developed primarily for calling mutations from solid tumor tissues. cfSNV can accurately detect mutations in cfDNA even with medium-coverage (e.g., ≥200×) sequencing, which makes whole-exome sequencing (WES) of cfDNA a viable option for various clinical utilities. Here, we present a user-friendly cfSNV package that exhibits fast computation and convenient user options. We also built a Docker image of it, which is designed to enable researchers and clinicians with a limited computational background to easily carry out analyses on both high-performance computing platforms and local computers. Mutation calling from a standard preprocessed WES dataset (~250× and ~70 million base pair target size) can be carried out in 3 h on a server with eight virtual CPUs and 32 GB of random access memory.

A bandwidth-aware service migration method in LEO satellite edge computing network

With the development of modern sensing technology, large amounts of data are generated in areas that lack terrestrial networks, such as the field and space. This data is transmitted by satellite networks, where the interesting data is only a small part. By introducing edge computing technology, identifying and transmitting interesting data will improve the efficiency of satellite network and the quality of service for users. A satellite edge computing (SatEC) system is proposed in this paper that provides computing services through LEO satellite network. For satellite characteristics with limited resources and mobility, a bandwidth-aware service migration problem is proposed to minimize the long-term total delay cost of the system while all tasks do not exceed its deadline. Converting long-term optimization problem into a series of one-time MINLP problems, an online lazy migration algorithm is firstly proposed to solve optimal service migration decision scheme with ϵ(1+1β)-competitive ratio. Then a hierarchical iterative algorithm is proposed to decouple communication and computing resources. Finally, a bandwidth-aware virtual CPU scheduling for post-copy migration algorithm is proposed to improve the migration speed. Simulation results show that the proposed scheme effectively deal with the topology change of satellite network and reduce the long-term total delay cost of the SatEC system.

Decentralized DNN Task Partitioning and Offloading Control in MEC Systems With Energy Harvesting Devices

In this paper, we study a decentralized deep neural network (DNN) task partitioning and offloading control problem for a multi-access edge computing (MEC) system with multiple wireless devices (WDs) powered by renewable energy sources. Instead of generic computational tasks, we focus on the case when the WDs and the MEC systems are computing DNN inferencing tasks. For each WD, we build a virtual local-computing CPU workload queue and an energy queue to model the DNN computation dynamics and the battery energy dynamics, respectively. We first introduce a new Lyapunov function based on the DNN tasks' cumulative execution latency and the total energy consumptions of the entire MEC system. Using a Lyapunov optimization approach, the DNN task partitioning and offloading control at the WDs are formulated as a Lyapunov drift minimization problem. To facilitate a distributed implementation, we exploit the linear structure of the cumulative latency and the independency of each WD's local-computing CPU rate, and we decompose the centralized Lyapunov drift minimization problem into multiple distributed subproblems, each of which is associated with a single WD. Next, we develop a parametric online learning algorithm such that each sub-Lyapunov drift minimization problem is solved locally at its associated WD. The proposed solution is completely decentralized in the sense that the sequential offloading and DNN task segmentation control at each WD is determined by its local battery energy queue state information (EQSI), the channel state information (CSI), and the DNN inferencing task workload queue state information (TQSI). Numerical results reveal that the proposed online learning algorithm can achieve significant performance gain over various state-of-the-art baselines.

Virtual Intermediate Bus CPU Voltage Regulator

This article presents a merged two-stage 48-V-to-1-V point-of-load (PoL) architecture with a 24-V virtual intermediate bus (VIB) for CPU voltage regulator applications. The VIB-PoL architecture includes two power conversion stages linked by a 24-V VIB with a significant voltage ripple. The first stage is a 2:1 interleaved charge pump, which converts 48 to 24 V. The second stage comprises multiple interleaved four-level series-capacitor buck modules with coupled inductors, converting 24 V to regulated 1 V with an equivalent voltage conversion ratio of 6:1. The VIB-PoL architecture achieves high efficiency and high power density by reducing the power conversion stress of both stages and eliminating the intermediate bus capacitors. A 48-V-to-1-V 640-A CPU voltage regulator with a peak power stage efficiency of 95.2% (93.3%, including gate driver loss), a full-load efficiency of 84.4% (83.1%, including gate driver loss), and a power density of 463 W/in <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^3$</tex-math></inline-formula> (at 1-V output with liquid cooling) is built and tested to demonstrate the VIB-PoL architecture.

Multimodal cloud resources utilization forecasting using a Bidirectional Gated Recurrent Unit predictor based on a power efficient Stacked denoising Autoencoders

To reap the advantages of many continual growing cloud services, cloud industries should adopt smart and holistic resources scheduling strategies. By deploying efficient deep learning technologies, many chaotic cloud traffics’ potential issues may be solved. Toward efficient cloud instances rightsizing and scheduling, we adopt in this paper a new Bidirectional Gated Recurrent Unit predictor based on a power efficient Stacked Denoising Autoencoders to forecast simultaneously future hourly virtual CPU, memory, and storage utilizations. Using various data ranges of resources under three AWS instances families, the best forecasting results achieved so far [1,83,30,78,331,11] of mean RMSE values and [1,37,21,63,245,13] of mean MAE values while predicting respectively future vCPU, memory, and storage utilizations. In addition, the proposed model also proved its precision stability and outperformance over the three considered SDAE-GRU, SDAE-LSTM and BiGRU benchmark models. Given the neglected power consumption measurement noticed in most related studies, we eventually validated the proposed predictor’s power efficiency by measuring in addition its real time consumed power in watt and temperature throughout the training process duration. The proposed predictor decreased the average consumed power by 5% compared to a classical sparse AE-BiGRU.

A wholistic optimization of containerized workflow scheduling and deployment in the cloud–edge environment

Digital Twin in Industry 4.0 utilizes Internet of Things (IoT) to collect real-life data and combine it with simulation models for product design and development. The simulation process can be executed as a workflow, consisting of tasks with precedence constraints. In a container-based workflow execution system, each task in the workflow is executed in a container within a virtual machine (VM). In this paper, a three-step scheduling model is proposed to combine scheduling of container-based workflows and the deployment of containers on a cloud–edge environment. In the first step, virtual CPU (vCPU) is allocated for each container to enable vCPU sharing among different containers. Next, two-step resource deployment is used to schedule the containers onto VM, and VM onto the physical machines in either edge or cloud environment. Multiple objectives are considered, including minimizing makespan, load imbalance, and energy consumption, from the perspective of cloud–edge resources as well as containerized workflows. To obtain a set of non-dominated solutions, three evolution strategies are designed and combined with two multi-objective algorithm frameworks — co-evolution strategy (CES), basic non-co-evolution strategy (B-NCS), and hybrid non-co-evolution strategy (H-NCS). Simulation results demonstrate that the proposed model outperforms the existing two-step scheduling model and H-NCS performs better than other strategies.

Fine-Grained I/O Traffic Control Middleware for I/O Fairness in Virtualized System

The development of IT technology in the 21st century has created a new paradigm for real-time, data-intensive user services, such as connected cars, smart factories, and remote health care services. The considerable computational resources required by these services are rendering the cloud increasingly more important. In the cloud server, user services are forced to share physical resources because of the emerging resource competition, thus introducing various types of unpredictable workloads. The core technology of the cloud is a virtualized system, which isolates and shares the powerful physical resources of the server in the form of a virtual machine (VM) to increase resource efficiency. However, the scheduling policy of a virtual CPU (vCPU), which is a logical CPU of a VM, generally schedules the vCPU based on the degree of occupation of the physical CPU (pCPU) without regarding I/O strength; so it brings the unfair I/O performance among VMs in the virtualized systems. The user services performing on the VM are not aware of the user-contention architectures, which sharing of I/O devices, in the virtualized systems; Furthermore, the current virtualized system simply adopts the Linux-based I/O processing process which optimized for user-contention-free architectures. Therefore, the architecture that brings the unfair usage of I/O devices among user services is hardly regarded and has low awareness in current virtualized systems. To overcome this problem, in this study, I-Balancer is presented to provide fair I/O performance among I/O-intensive user services by applying an asynchronous inter-communication control technique for the virtualized system with a high VM density. The main design goal of I-Balancer is to increase the awareness of user-contention architectures in the hypervisor. I-Balancer derives the fine-grained workload and I/O strength for each vCPU during the scheduler and event channel areas. Subsequently, to strengthen fair I/O performance, an I/O traffic control mechanism is implemented to control the inter-domain communication traffic according to the I/O strength of the VMs. The experimental results prove that I-Balancer provides a fairer network and disk I/O performance than the existing virtualized systems with negligible performance interference and overhead.

Proactive load balancing fault tolerance algorithm in cloud computing

AbstractAs the demand for cloud computing infrastructure increased, availability and reliability have become more important as they are major features in real‐time computing systems. The chances of failure in a cloud computing system increase compared to other computing systems. To achieve high reliability and availability for the cloud computing infrastructure fault tolerance is one of the most essential modules. For a cloud computing infrastructure, a proactive fault tolerance algorithm with load balancing is proposed in this paper. The proposed methodology tolerates virtual machine CPU faults to achieve maximum cloud computing infrastructure reliability and availability. CPU faults can occur during virtual machine operation. The proposed model's main objective is to track changes in CPU utilization and make a decision once a high value of CPU utilization is identified. The decision is either to migrate the defective virtual machine to another “destination host”, or to manage loads on the destination host so that the faulty virtual machine can be migrated. A load is calculated for the destination host; if the destination host is overloaded, the subroutine for load balancing is launched. Load balancing is based on the technique of choosing a virtual machine.

Predictive Scaling for Elastic Compute Resources on Public Cloud Utilizing Deep Learning based Long Short-term Memory

The cloud resource usage has been increased exponentially because of adaptation of digitalization in government and corporate organization. This might increase the usage of cloud compute instances, resulting in massive consumption of energy from High performance Public Cloud Data Center servers. In cloud, there are some web applications which may experience diverse workloads at different timestamps that are essential for workload efficiency as well as feasibility of all extent. In cloud application, one of the major features is scalability in which most Cloud Service Providers (CSP) offer Infrastructure as a Service (IaaS) and have implemented auto-scaling on the Virtual Machine (VM) levels. Auto-scaling is a cloud computing feature which has the ability in scaling the resources based on demand and it assists in providing better results for other features like high availability, fault tolerance, energy efficiency, cost management, etc. In the existing approach, the reactive scaling with fixed or smart static threshold do not fulfill the requirement of application to run without hurdles during peak workloads, however this paper focuses on increasing the green tracing over cloud computing through proposed approach using predictive auto-scaling technique for reducing over-provisioning or under-provisioning of instances with history of traces. On the other hand, it offers right sized instances that fit the application to execute in satisfying the users through on-demand with elasticity. This can be done using Deep Learning based Time-Series LSTM Networks, wherein the virtual CPU core instances can be accurately scaled using cool visualization insights after the model has been trained. Moreover, the LSTM accuracy result of prediction is also compared with Gated Recurrent Unit (GRU) to bring business intelligence through analytics with reduced energy, cost and environmental sustainability.

I/O Strength-Aware Credit Scheduler for Virtualized Environments

With the evolution of cloud technology, the number of user applications is increasing, and computational workloads are becoming increasingly diverse and unpredictable. However, cloud data centers still exhibit a low I/O performance because of the scheduling policies employed, which are based on the degree of physical CPU (pCPU) occupancy. Notably, existing scheduling policies cannot guarantee good I/O performance because of the uncertainty of the extent of I/O occurrence and the lack of fine-grained workload classification. To overcome these limitations, we propose ISACS, an I/O strength-aware credit scheduler for virtualized environments. Based on the Credit2 scheduler, ISACS provides a fine-grained workload-aware scheduling technique to mitigate I/O performance degradation in virtualized environments. Further, ISACS uses the event channel mechanism in the virtualization architecture to expand the scope of the scheduling information area and measures the I/O strength of each virtual CPU (vCPU) in the run-queue. Then, ISACS allocates two types of virtual credits for all vCPUs in the run-queue to increase I/O performance and concurrently prevent CPU performance degradation. Finally, through I/O load balancing, ISACS prevents I/O-intensive vCPUs from becoming concentrated on specific cores. Our experiments show that compared with existing virtualization environments, ISACS provides a higher I/O performance with a negligible impact on CPU performance.

Reconciling Time Slice Conflicts of Virtual Machines With Dual Time Slice for Clouds

The proliferation of system virtualization poses a new challenge for the coarse-grained time sharing techniques for consolidation, since operating systems are running on virtual CPUs. The current system stack was designed under the assumption that operating systems can seize CPU resources at any moment. However, for the guest operating system on a virtual machine (VM), such assumption cannot be guaranteed, since virtual CPUs of VMs share a limited number of physical cores. Due to the time-sharing of physical cores, the execution of a virtual CPU is not contiguous, with a gap between the virtual and real time spaces. Such a virtual time discontinuity problem leads to significant inefficiency for lock and interrupt handling, which rely on the immediate availability of CPUs whenever the operating system requires computation. To reduce scheduling latencies of virtual CPUs, shortening time slices can be a straightforward strategy, but it may lead to the increased overhead of context switching costs across virtual machines for some workloads. It is challenging to determine a single time slice to satisfy all the VMs. In this article, we propose to have dual time slice to resolve the time slice conflict problem occurred in different types of virtual machines.

Study of Virtual Network Function Placement in 5G Cloud Radio Access Network

5G and beyond need to meet stringent requirements of latency, reliability, and support for heterogeneous devices. However, the existing wireless network architecture is limited to fulfill these constraints. Cloud radio access network, along with network function virtualization, is suggested to provide flexibility and network agility. It decouples network functions, such as firewall and packet gateway, from hardware to software deployed in the cloud. Thus comprehensive end-to-end formulation of this architecture is required for virtual network function placement. Most of existing works focus on virtual functions placement with different objectives, addressing different service requirements separately. In this article, six 5G constraints are considered simultaneously to find optimal virtual network function placement with service differentiation. The selected six parameters reflect services’ requirements, network constraints and computing constraints. We first model the overall cloud radio access network as a multi-layer loopless-random hypergraph and we provide the overall formulation of the system. Then, we reformulate such model considering backup virtual functions and CPU over-provisioning techniques to improve both virtual function’s reliability and processing latency. Finally, we propose service differentiation to reduce CPU utilization and energy consumption, while using the above techniques. The results suggest that the application of service differentiation can significantly improve assignment of computing resources and energy efficiency.

Dynamic Consolidation of Virtual Machine: A Survey of Challenges for Resource Optimization in Cloud Computing

Background:: Virtualization is an efficient technology that accelerates available data center to support efficient workload for the application. It completely based on guest operating system which keeps track of infrastructure that keeps track of real time usage of hardware and utilization of software. Objective:: To address the issues with Virtualization this paper analyzed various virtualization terminology for treating best effective way to reduce IT expenses while boosting efficiency and deployment for all levels of businesses. Methods: This paper discusses about the scenarios where various challenges met by Dynamic VM consolidation. Dynamic conclusion of virtual machines has the ability to increase the consumption of physical setup and focus on reducing power utilization with VM movement for stipulated period. Gathering the needs of all VM working in the application, adjusting the Virtual machine and suitably fit the virtual resource on a physical machine. Profiling and scheduling the virtual CPU to another Physical resource. This can be increased by making live migration with regards to planned schedule of virtual machine allotment. Results:: The recent trends followed in comprehending dynamic VM consolidation is applicable either in heuristic-based techniques which has further approaches based on static as well as adaptive utilization threshold. SLA with unit of time with variant HOST adoption (SLATAH) which is dependent on CPU utilization threshold with 100% for active host. Conclusion:: The cloud provider decision upon choosing the virtual machine for their application also varies with their decision support system that considers data storage and other parameters. It is being compared for the continuous workload distribution as well as eventually compared with changing demands of computation and in various optimization VM placement strategies.

Virtual CPU Scaling for Efficient Server Power Consumption in Cloud Data Center

Cloud computing is a fast-growing technology used by major corporations these days because of the flexibility framework it provides to consumers. Cloud technology requires large data centers consisting of multiple IT equipment and servers. One main problem with these data centers is the vast amount of power consumed during servers operation. This reduces financial benefit and increases the need to produce more energy to cover the needs of operating the cloud infrastructure. This paper proposes an approach for managing the virtual central processing unit (vCPU) of a virtual machine to improve server power efficiency. A framework is used to study the proposed approach while processing different types of workloads widely found in most general-purpose cloud computing applications. Results indicate an improvement in server power saving.

VCSP: virtual CPU scheduling for post-copy live migration of virtual machines

The live migration of virtual machines among physical machines aims at efficient utilization of resources, load balancing, maintenance, energy management, fault tolerance, sharing resources and mobile computing. There are several methods for the live migration of virtual machines. In the post-copy approach, a virtual machine starts working in a target host, when data that is not present in the target machine is requested, a page fault occurs and the processing stops in the target until the arrival of the requested information. Certainly, the number of stops and lags negatively affects the system downtime. The reduction of physical or virtual CPU frequency is one of the techniques recommended to efficient migration of virtual machines. The modification of the frequency of physical and/or virtual CPU has already been used in the improvement of the pre-copy method to manage the speed of changes in the source machine. This paper presents the virtual CPU scheduling for post-copy (VSCP) framework that reduces the speed of the virtual CPU in post-copy to make a balance between processing speed of pages in the target machine and their transmission speed in the source machine. The VCSP is also compared with the baseline methods of the post-copy-prefetching and post-copy. In the experiments, the system downtime, total migration time, total pages transferred and throughput of system are evaluated. The results indicate that the proposed method improves the system downtime up to 8.17%, total migration time up to 30.33%, total pages transferred up to 54.49% and throughput of system up to 23.65%.

Resource allocation, scheduling and auto-scaling algorithms for enhancing the performance of cloud using Grey Wolf Optimization and Fuzzy rules

Cloud computing technology is playing a major role in the industry and real-life, for providing fast services such as data sharing and allocating the cloud resources that are paid and truly required. In this scenario, the cloud users are scheduled according to the rule-based systems for attempting to automate the matching between computing requirements and resources. Even though, the majority auto-scaling algorithms only helped as indicators for simple resource utilization and also not considered both cloud user needs and budget concerns. For this purpose, we propose a new model which is the combination of auto-scaling algorithms, resource allocation and scheduling for allocating the appropriate resources and scheduled them. This model consists of three new algorithms namely Grey Wolf Optimization and Fuzzy rules based Resource allocation and Scheduling Algorithm (GWOFRSA), Auto-Scaling Algorithm for Cloud based Web Application (ASACWA) and Auto-Scaling Algorithm for handling Distributed Computing Tasks (ASADCT). Here, we introduce new auto-scaling algorithms for enhancing the performance of cloud services. In this work, the optimization technique is used to predict the cloud server workload, resource requirements and it also uses fuzzy rules for monitoring the resource utilization and the size of virtual machine allocation process. According to the workload prediction, the completion time is estimated for each cloud server. The experiments are conducted by using a simulator called CloudSim environment of Java programming and compared with the existing works available in this direction in terms of resource utilization and enhance the cloud performance with better Quality of Service of Virtual Machine allocation, Missed Deadline, Demand Satisfaction, Power Utilization, CPU Load and throughput.

Virtual Cluster Deployment Model for Large-Scale Data Processing Jobs

The virtual cluster needs to deploy virtual nodes with different computing modes for Large-scale data computing jobs including many tasks with different computing modes. There are resource unbalance among the virtual cluster subsets, which is consisted of virtual nodes with the same computing mode. So the resource utilization of virtual cluster is poor due to the time-varying and complexity of workloads. To solve these issues, a novel dynamic deployment model based on Docker was proposed to optimize resource configuration of virtual cluster. Hybrid deployment scheme includes coarse-grained deployment and fine-grained deployment in order to absorb the mutation of resource demand when processing the jobs encapsulated with complex computing logic. This deployment model can reduce job waiting time and improve performance by providing more resource for running jobs. The coarse-grained deployment mechanism has the ability of decreasing deploying overhead and improving stability from global optimal perspective. And the fine-grained deployment mechanism can increases the deploying accuracy from local optimal perspective. The experiments show that this model can improve the execution efficiency of job by 2.3 percent compared with static deploying scheme and improve the virtual CPU utilization about 7.9 to 25.9 percent compared with other approaches.