Self-managing cloud-native applications: Design,implementation, and experience

Running applications in the cloud efficiently requires much more than deploying software in virtual machines. Cloud applications have to be continuously managed: (1) to adjust their resources to the incoming load and (2) to face transient failures replicating and restarting components to provide resiliency on unreliable infrastructure. Continuous management monitors application and infrastructural metrics to provide automated and responsive reactions to failures (health management) and changing environmental conditions (auto-scaling) minimizing human intervention.In the current practice, management functionalities are provided as infrastructural or third party services. In both cases they are external to the application deployment. We claim that this approach has intrinsic limits, namely that separating management functionalities from the application prevents them from naturally scaling with the application and requires additional management code and human intervention. Moreover, using infrastructure provider services for management functionalities results in vendor lock-in effectively preventing cloud applications to adapt and run on the most effective cloud for the job.In this paper we discuss the main characteristics of cloud native applications, propose a novel architecture that enables scalable and resilient self-managing applications in the cloud, and relate on our experience in porting a legacy application to the cloud applying cloud-native principles.     

ICENI: Optimisation of component applications within a Grid environment

Effective exploitation of Computational Grids can only be achieved when applications are fully integrated with the Grid middleware and the underlying computational resources. Fundamental to this exploitation is information. Information about the structure and behaviour of the application, the capability of the computational and networking resources, and the availability and access to these resources by an individual, a group or an organisation.

In this paper we describe Imperial College e-Science Networked Infrastructure (ICENI), a Grid middleware framework developed within the London e-Science Centre. ICENI is a platform-independent framework that uses open and extensible XML derived protocols, within a framework built using Java and Jini, to explore effective application execution upon distributed federated resources. We match a high-level application specification, defined as a network of components, to an optimal combination of the currently available component implementations within our Grid environment, by using composite performance models. We demonstrate the effectiveness of this architecture through the high-level specification and solution of a set of linear equations by automatic and selection of optimal resources and implementations.     

On minimizing the resource consumption of cloud applications using process migrations

According to the pay-per-use model adopted in clouds, the more resources an application running in a cloud computing environment consumes, the greater the amount of money the owner of the corresponding application will be charged. Therefore, applying intelligent solutions to minimize the resource consumption is of great importance. In this paper, we study the problem of identifying an assignment scheme between the interacting components of an application, such as processes and virtual machines, and the computing nodes of a cloud system, such that the total amount of resources consumed by the respective application is minimized. Because centralized solutions are deemed unsuitable for large distributed systems or large-scale applications, we propose a fully distributed algorithm (called DRA) to overcome scalability issues. DRA takes decisions concerning the transition from one assignment scheme to another in a dynamic way, based solely on local information. We also propose and test two modifications of the basic DRA algorithm to deal better with the heterogeneity of cloud servers in terms of capacity constraints. We must note that we capture heterogeneity regarding the network model. Through theoretical analysis, we formally prove that DRA achieves convergence and always provides an optimal solution for tree-based networks in the uncapacitated case. Moreover, we prove through experimental evaluation that DRA achieves up to 55% network cost reduction when compared to the most recent algorithm in the literature. We also show that the proposed modifications of DRA improve the algorithm’s performance considerably in the case where servers have limited capacity.     

公有云存储系统性能评测方法研究

随着云存储系统的迅速发展和广泛使用,许多企业开发者和个人用户将其应用从传统存储迁移至公有云存储系统,因此,云存储系统性能成为企业开发者和个人用户关注的焦点。由于传统测试难以模拟足够多的用户同时访问云存储系统;测试环境构建复杂,测试时间长,准备测试环境成本高;受网络因素及外界其他因素影响,评测结果不稳定。针对以上所述云存储系统性能评测的重点和难点,提出一种“云测试云”的公有云存储系统性能评测方法,该方法通过在云计算平台动态申请足够数量的实例,对公有云存储系统性能进行评测。首先,构建通用的性能评测框架,可动态伸缩申请实例,自动化部署评测工具及负载,控制并发访问云存储系统,自动释放实例及收集并反馈评测结果;其次,提出多维度的性能评测指标,涵盖不同典型应用、不同云存储接口;最后,提出一种可扩展通用的性能评测模型,该模型可以评测常见典型应用的性能,分析云存储性能影响因素,可适用于任何的公有云存储平台。为了验证该方法的可行性、合理性、通用性和可扩展性,利用所提方法对Amazon S3云存储系统进行性能评测,并使用s3cmd验证评测结果的准确性。实验结果表明,评测结果可以为企业开发者和个人用户提供参考意见。     

Implementation and performance evaluation of a scheduling algorithm for divisible load parallel applications in a cloud computing environment

Cloud computing is an emerging technology in which information technology resources are virtualized to users in a set of computing resources on a pay‐per‐use basis. It is seen as an effective infrastructure for high performance applications. Divisible load applications occur in many scientific and engineering applications. However, dividing an application and deploying it in a cloud computing environment face challenges to obtain an optimal performance due to the overheads introduced by the cloud virtualization and the supporting cloud middleware. Therefore, we provide results of series of extensive experiments in scheduling divisible load application in a Cloud environment to decrease the overall application execution time considering the cloud networking and computing capacities presented to the application's user. We experiment with real applications within the Amazon cloud computing environment. Our extensive experiments analyze the reasons of the discrepancies between a theoretical model and the reality and propose adequate solutions. These discrepancies are due to three factors: the network behavior, the application behavior and the cloud computing virtualization. Our results show that applying the algorithm result in a maximum ratio of 1.41 of the measured normalized makespan versus the ideal makespan for application in which the communication to computation ratio is big. They show that the algorithm is effective for those applications in a heterogeneous setting reaching a ratio of 1.28 for large data sets. For application following the ensemble clustering model in which the computation to communication ratio is big and variable, we obtained a maximum ratio of 4.7 for large data set and a ratio of 2.11 for small data set. Applying the algorithm also results in an important speedup. These results are revealing for the type of applications we consider under experiments. The experiments also reveal the impact of the choice of the platforms provided by Amazon on the performance of the applications under study. Considering the emergence of cloud computing for high performance applications, the results in this paper can be widely adopted by cloud computing developers. Copyright © 2014 John Wiley & Sons, Ltd.     

A flow measurement architecture to preserve application structure

The Internet has significantly evolved in the number and variety of applications. Network operators need mechanisms to constantly monitor and study these applications. Modern routers employ passive measurement solution called Sampled NetFlow to collect basic statistics on a per-flow basis (for a small subset of flows), that could provide valuable information for application monitoring. Given modern applications routinely consist of several flows, potentially to many different destinations, only a few flows are sampled per application session using Sampled NetFlow. To address this issue, in this paper, we introduce related sampling that allows network operators to give a higher probability to flows that are part of the same application session. Given the lack of application semantics in the middle of the network, our architecture, RelSamp, treats flows that share the same source IP address as related. Our heuristic works well in practice as hosts typically run few applications at any given instant, as observed using a measurement study on real traces. In our evaluation using real traces, we show that RelSamp achieves 5–10× more flows per application session compared to Sampled NetFlow for the same effective number of sampled packets. We also show that behavioral and statistical classification approaches such as BLINC, SVM and C4.5 achieve up to 50% better classification accuracy compared to Sampled NetFlow, while not impairing existing management tasks such as volume estimation too much.     

Latency based group discovery algorithm for network aware cloud scheduling

Cloud computing is a big paradigm shift of computing mechanism. It provides high scalability and elasticity with a range of on-demand services. We can execute a variety of distributed applications on cloud’s virtual machines (computing nodes). In a distributed application, virtual machine nodes need to communicate and coordinate with each other. This type of coordination requires that the inter-node latency should be minimal to improve the performance. But in the case of nodes belonging to different clusters of the same cloud or in a multi-cloud environment, there can be a problem of higher network latency. So it becomes more difficult to decide, which node(s) to choose for the distributed application execution, to keep inter-node latency at minimum. In this paper, we propose a solution for this problem. We propose a model for the grouping of nodes with respect to network latency. The application scheduling is done on the basis of network latency. This model is a part of our proposed Cloud Scheduler module, which helps the scheduler in scheduling decisions on the basis of different criteria. Network latency and resultant node grouping on the basis of this latency is one of those criteria. The main essence of the paper is that our proposed latency grouping algorithm not only has no additional network traffic overheads for algorithm computation but also works well with incomplete latency information and performs intelligent grouping on the basis of latency. This paper addresses an important problem in cloud computing, which is locating communicating virtual machines for minimum latency between them and group them with respect to inter-node latency.     

CloudRank-D: benchmarking and ranking cloud computing systems for data processing applications

With the explosive growth of information, more and more organizations are deploying private cloud systems or renting public cloud systems to process big data. However, there is no existing benchmark suite for evaluating cloud performance on the whole system level. To the best of our knowledge, this paper proposes the first benchmark suite CloudRank-D to benchmark and rank cloud computing systems that are shared for running big data applications. We analyze the limitations of previous metrics, e.g., floating point operations, for evaluating a cloud computing system, and propose two simple metrics: data processed per second and data processed per Joule as two complementary metrics for evaluating cloud computing systems. We detail the design of CloudRank-D that considers representative applications, diversity of data characteristics, and dynamic behaviors of both applications and system software platforms. Through experiments, we demonstrate the advantages of our proposed metrics. In several case studies, we evaluate two small-scale deployments of cloud computing systems using CloudRank-D.     

Dynamic demultiplexing in network calculus—Theory and application

During the last two decades, starting with the seminal work by Cruz, network calculus has evolved as a new theory for the performance analysis of networked systems. In contrast to classical queueing theory, it deals with performance bounds instead of average values and thus has been the theoretical basis of quality of service proposals such as the IETF’s Integrated and Differentiated Services architectures. Besides these it has, however, recently seen many other application scenarios as, for example, wireless sensor networks, switched Ethernets, avionic networks, Systems-on-Chip, or even to speed-up simulations, to name a few.In this article, we extend network calculus by adding a new versatile modeling element: a demultiplexer. Conventionally, demultiplexing has been either neglected or assumed to be static, i.e., fixed at the setup time of a network. This is restrictive for many potential applications of network calculus. For example, a load balancing based on current link loads in a network could not be modeled with conventional network calculus means. Our demultiplexing element is based on stochastic scaling. Stochastic scaling allows one to put probabilistic bounds on how a flow is split inside the network. Fundamental results on network calculus with stochastic scaling are therefore derived in this work. We illustrate the benefits of the demultiplexer in a sample application of uncertain load balancing.     

Simulation of SLA-based VM-scaling algorithms for cloud-distributed applications

Cloud Computing has evolved to become an enabler for delivering access to large scale distributed applications running on managed network-connected computing systems. This makes possible hosting Distributed Enterprise Information Systems (dEISs) in cloud environments, while enforcing strict performance and quality of service requirements, defined using Service Level Agreements (SLAs). SLAs define the performance boundaries of distributed applications, and are enforced by a cloud management system (CMS) dynamically allocating the available computing resources to the cloud services. We present two novel VM-scaling algorithms focused on dEIS systems, which optimally detect most appropriate scaling conditions using performance-models of distributed applications derived from constant-workload benchmarks, together with SLA-specified performance constraints. We simulate the VM-scaling algorithms in a cloud simulator and compare against trace-based performance models of dEISs. We compare a total of three SLA-based VM-scaling algorithms (one using prediction mechanisms) based on a real-world application scenario involving a large variable number of users. Our results show that it is beneficial to use autoregressive predictive SLA-driven scaling algorithms in cloud management systems for guaranteeing performance invariants of distributed cloud applications, as opposed to using only reactive SLA-based VM-scaling algorithms.     

基于粗糙集的容器云系统健康度评价建模

现在容器云平台容器数目日益增加,相关监控数据爆炸式增长,而现有的运行在容器内的微服务监控软件监控指标不仅种类繁多,配置繁琐,并且往往只是直接给出监控数据,没有根据得到的监控指标对系统的健康度进行度量。针对该问题,提出了一种新的基于粗糙集的容器云系统健康度评价模型。通过建立的粗糙集云系统健康度评价模型,可以直观地反映整个集群的健康程度。首先通过信息熵对监控到的连续属性进行断点分割,离散化处理,然后利用粗糙集理论实现对监控数据进行知识约简、一致性检查和决策表建立,从而建立了基于粗糙集和信息熵的集群健康度指标模型。最后,通过Kubernetes容器云平台分别进行计算密集负载和网络密集负载仿真实验,实验结果表明,该模型能够反映集群的性能和对异常进行检测。     

An intelligent memory caching architecture for data-intensive multimedia applications

With the rapid developments in cloud computing and mobile networks, multimedia content can be accessed conveniently. Recently, some novel intelligent caching-based approaches have been proposed to improve the memory architectures for multimedia applications. These applications often face bottleneck related challenges which result in performance degradation and service delay issues. Intelligent multimedia network applications access the shared data by using a specific network file system. This results in answering the processing related constraints on hard-drive storage and might result in bringing bottleneck issues. Therefore, to improve the performance of these multimedia network applications, we present an intelligent distributed memory caching system. We integrate the multimedia application message passing interface in a multi-threaded environment and propose an algorithm which can handle concurrent response behavior for different multimedia applications. Results demonstrate that our proposed scheme outperforms traditional approaches in terms of throughput and file read access features.

     

Topology and Application Aware Dynamic VM Management in the Cloud

Cloud computing continues to mature and more applications continue to be deployed in public clouds. Client applications deployed in the cloud should automatically scale up and down to match changing workload demands, though they must be careful to ensure that sufficient resources are provisioned to achieve performance objectives. The cloud provider, on the other hand, attempts to reduce costs by reducing power consumption by consolidating load onto fewer, highly utilized machines. In this work, we introduce an algorithm that integrates both application autoscaling and dynamic virtual machine (VM) allocation into a single algorithm in order to achieve the goals of both cloud provider and client. Further, we consider multi-VM applications, such as multi-tiered web-based applications, and extend the integrated algorithm to take the network topology into account when placing or migrating applications. The goal is to reduce VM-to-VM communication latency; our focus is on trying to contain applications within the same racks. We evaluate our work through simulation, showing that the integrated algorithm can achieve better application performance with a significant reduction in virtual machine live migrations, and the topology-aware extension successfully places applications within a single rack.     

Parallel patterns for heterogeneous CPU/GPU architectures: Structured parallelism from cluster to cloud

The widespread adoption of traditional heterogeneous systems has substantially improved the computing power available and, in the meantime, raised optimisation issues related to the processing of task streams across both CPU and GPU cores in heterogeneous systems. Similar to the heterogeneous improvement gained in traditional systems, cloud computing has started to add heterogeneity support, typically through GPU instances, to the conventional CPU-based cloud resources. This optimisation of cloud resources will arguably have a real impact when running on-demand computationally-intensive applications.In this work, we investigate the scaling of pattern-based parallel applications from physical, “local” mixed CPU/GPU-clusters to a public cloud CPU/GPU infrastructure. Specifically, such parallel patterns are deployed via algorithmic skeletons to exploit a peculiar parallel behaviour while hiding implementation details.We propose a systematic methodology to exploit approximated analytical performance/cost models, and an integrated programming framework that is suitable for targeting both local and remote resources to support the offloading of computations from structured parallel applications to heterogeneous cloud resources, such that performance values not available on local resources may be actually achieved with the remote resources. The amount of remote resources necessary to achieve a given performance target is calculated through the performance models in order to allow any user to hire the amount of cloud resources needed to achieve a given target performance value. Thus, it is therefore expected that such models can be used to devise the optimal proportion of computations to be allocated on different remote nodes for Big Data computations.We present different experiments run with a proof-of-concept implementation based on FastFlow  on small departmental clusters as well as on a public cloud infrastructure with CPU and GPU using the Amazon Elastic Compute Cloud. In particular, we show how CPU-only and mixed CPU/GPU computations can be offloaded to remote cloud resources with predictable performances and how data intensive applications can be mapped to a mix of local and remote resources to guarantee optimal performances.     

面向5G网络云原生应用资源调度的博弈优化策略

随着5G网络和云原生技术的发展,面向服务的5G云原生核心网应运而生,传统应用正朝着云原生化方向发展。目前云原生服务提供商和云原生应用商数量众多且关系复杂,使得应用在云原生化过程中的资源调度面临新挑战。提出一种5G网络云原生应用资源调度优化策略,将云原生应用商和云原生服务提供商构建为多主多从的Stackelberg博弈模型,对传统收益进行具体描述并联合能耗构建利润函数和策略空间,证明给定一组微服务资源定价的情况下存在云原生应用商的纳什均衡点。在此基础上,引入柯西分布对策略进行优化,提高其收敛性能,通过分布式迭代方法得到云原生服务提供商的最佳微服务定价和云原生应用商的最佳微服务租用比例。仿真结果表明,相比ACA算法、QOS PA算法以及GOS策略,该策略能够有效提高网络收益和用户体验质量,同时降低应用开发能耗。     

A real-time network simulation infrastructure based on OpenVPN

We present an open and flexible software infrastructure that embeds physical hosts in a simulated network. In real-time network simulation, where real-world implementations of distributed applications and network services can run together with the network simulator that operates in real-time, real network packets are injected into the simulation system and subject to the simulated network conditions computed as a result of both real and virtual traffic traversing the network and competing for network resources. Our real-time simulation infrastructure has been implemented based on Open Virtual Private Network (OpenVPN), modified and customized to bridges traffic between the physical hosts and the simulated network. We identify the performance advantages and limitations of our approach via a set of experiments. We also present two interesting application scenarios to show the capabilities of the real-time simulation infrastructure.     

Enabling fuzzy technologies in high performance networking via an open FPGA-based development platform

Soft computing techniques and particularly fuzzy inference systems are gaining momentum as tools for network traffic modeling, analysis and control. Efficient hardware implementations of these techniques that can achieve real-time operation in high-speed networking equipment as well as other highly time-constrained application fields is however an open problem. We introduce a development platform for fuzzy inference systems with applications to network traffic analysis and control. The platform addresses the current requirements and constraints of high performance networking equipment. For the development process, we set up a methodology and a CAD tool chain that span the entire design process from initial specification in a high-level language to implementation on FPGA devices. An FPGA development board with PCI/PCIe interface is employed to support an open platform that comprises CAD tools as well as IP cores. PCI compatible fuzzy inference modules are implemented as System-on-Programmable-Chip (SoPC). We present satisfactory experimental results from the implementation of fuzzy systems for a number of applications in analysis and control of Internet traffic. These systems are shown to satisfy operational and architectural requirements of current and future high performance routing equipment. The platform proposed allows for the development of prototypes while avoiding large investments and complicated management procedures which constrain the testing and adoption of soft computing techniques in high performance networking.     

Cloud computing in e-Science: research challenges and opportunities

Service-oriented architecture (SOA), workflow, the Semantic Web, and Grid computing are key enabling information technologies in the development of increasingly sophisticated e-Science infrastructures and application platforms. While the emergence of Cloud computing as a new computing paradigm has provided new directions and opportunities for e-Science infrastructure development, it also presents some challenges. Scientific research is increasingly finding that it is difficult to handle “big data” using traditional data processing techniques. Such challenges demonstrate the need for a comprehensive analysis on using the above-mentioned informatics techniques to develop appropriate e-Science infrastructure and platforms in the context of Cloud computing. This survey paper describes recent research advances in applying informatics techniques to facilitate scientific research particularly from the Cloud computing perspective. Our particular contributions include identifying associated research challenges and opportunities, presenting lessons learned, and describing our future vision for applying Cloud computing to e-Science. We believe our research findings can help indicate the future trend of e-Science, and can inform funding and research directions in how to more appropriately employ computing technologies in scientific research. We point out the open research issues hoping to spark new development and innovation in the e-Science field.