Public Cloud Resources Research Articles

Efficient I/O Performance-Focused Scheduling in High-Performance Computing

High-performance computing (HPC) systems are becoming increasingly important as contemporary exascale applications with demand extensive computational and data processing capability. To optimize these systems, efficient scheduling of HPC applications is important. In particular, because I/O is a shared resource among applications and is becoming more important due to the emergence of big data, it is possible to improve performance by considering the architecture of HPC systems and scheduling jobs based on I/O resource requirements. In this paper, we propose a scheduling scheme that prioritizes HPC applications based on their I/O requirements. To accomplish this, our scheme analyzes the IOPS of scheduled applications by examining their execution history. Then, it schedules the applications at pre-configured intervals based on their expected IOPS to maximize the available IOPS across the entire system. Compared to the existing first-come first-served (FCFS) algorithm, experimental results using real-world HPC log data show that our scheme reduces total execution time by 305 h and decreases costs by USD 53 when scheduling 10,000 jobs utilizing public cloud resources.

From raw seismic data to quantitative interpretation via cloud-based FWI

Full-waveform inversion (FWI) has evolved in recent years from a technique for recovering long- to intermediate-wavelength updates of acoustic overburden velocities as part of a broader model building workflow to a standalone tool for high-fidelity seismic imaging using raw seismic data – with ever greater resolution and ever more sophisticated physics being sought. In doing so, it has moved notably closer to fulfilling the vision of its original inventors in the late 70s and early 80s. In this work, two novel approaches to FWI along with enabling technologies for their effective deployment are considered. The first of these allows for robust long-wavelength updates to be generated from raw seismic data in the absence of diving wave energy, providing reliable estimates of acoustic velocity at depths equivalent to or greater than the longest offsets available in the data – thereby enabling more effective imaging with data of limited quality. The second allows for the estimation of elastic amplitude-versus-angle behaviour while still using an acoustic wave equation, thereby mitigating the significant computational burden typically associated with elastic FWI. Finally, it is demonstrated that these approaches can be deployed in a reliable and cost-effective manner via the considered use of public cloud resources, with learnings relevant for analogous massively parallel scientific applications. Case studies from two marine environments are provided for discussion, one being from deep-water US Gulf of Mexico and the other from intermediate water depths offshore Western Australia.

Cloud Architectures for Distributed Multi-Cloud Computing: A Review of Hybrid and Federated Cloud Environment

The concept of several clouds has greatly extended the use of cloud computing and gained popularity in academic and business circles. The use of multi-cloud techniques has increased as businesses use cloud computing more and more to meet their computational demands. A thorough analysis of cloud architectures intended for distributed multi-cloud computing is presented in this study, with an emphasis on federated and hybrid cloud systems. The study looks at the opportunities and difficulties of adopting and overseeing a variety of cloud resources from several providers. The review starts out by going over the basic ideas and reasons for using multi-cloud strategies, emphasizing how important flexibility, scalability, and resilience are in contemporary computing settings. The study then explores the nuances of hybrid cloud architectures, with a focus on how private and public cloud resources can be seamlessly combined. In the context of hybrid cloud installations, important factors including data sovereignty, security, and workload orchestration are covered. In addition, the research delves into federated cloud architectures, clarifying how enterprises can coordinate and oversee workloads across several cloud providers. An examination of resource identification, policy enforcement, and interoperability procedures sheds light on the intricacies of federated cloud computing. The review delves into new developments in standards, best practices, and technology that help multi-cloud ecosystems mature. The study analyses the state of research and industry practices now, pointing out gaps and possible directions for future development. The intention is to provide decision-makers, researchers, and practitioners with a comprehensive grasp of the changing cloud architectural scene so they can plan and execute distributed multi-cloud solutions with knowledge. In conclusion, this article provides a thorough overview of hybrid and federated cloud architectures by combining information from many sources. Through a comprehensive analysis of the difficulties and possibilities associated with multi-cloud computing, the study hopes to add to the current conversation on cloud environment design and optimization in the rapidly changing technological landscape.

Serverless application composition leveraging function fusion: Theory and algorithms

Serverless computing is a novel cloud computing paradigm enabling flexible, cost-efficient and fine-granular development of cloud-native applications, although without any guarantees on scheduling or execution times. Thus, various high-level solutions have been proposed in recent years to find proper configurations of individual FaaS functions, while considering user-given QoS requirements. However, the ever-increasing complexity of invocation patterns among stateless functions, externalized management of intermediate states, and diverse public cloud resources pose new challenges to the composition of highly data-intensive serverless applications. In this paper, we fill this gap by proposing novel algorithms based on the emerging function fusion technique, along with the related cost/performance models of composite functions supporting implicit instance parallelization and internal state propagation. We prove the NP-completeness of the underlying latency-constrained tree partitioning problem, and design a bicriteria approximation scheme and a greedy heuristic to derive cost-efficient deployment configurations in polynomial time. With the help of extensive simulations using synthetic call graphs generated from public cloud traces, we demonstrate the applicability and superior runtime performance of our proposed methods compared to state-of-the-art solutions. In addition, we showcase that further cost-reduction of up to 3–6 % can be achieved compared to the optimal partitioning with the allowance of tolerable latency violations.

Enabling the execution of HPC applications on public clouds with HPC@Cloud toolkit

AbstractThe advent of cloud computing has made access to computing infrastructure available to millions of users that face resource constraints. In the context of high performance computing (HPC), public cloud resources have emerged as a cost‐effective alternative to expensive on‐premises clusters. However, there are several challenges and limitations in adopting this approach. This paper proposes HPC@Cloud , a provider‐agnostic open‐source software toolkit that facilitates the migration, testing, and execution of HPC applications in public clouds. The toolkit takes advantage of various fault tolerance technologies to enable the use of inexpensive transient cloud infrastructure, commonly known as “spot” instances. Also, it features integration with singularity containers, allowing users to run complex applications on virtual HPC clusters in a portable and reproducible way. Finally, it provides a data‐based empirical approach to estimating cloud infrastructure costs for HPC workloads. The results obtained on two public cloud providers (AWS and Vultr) show that: (i) HPC@Cloud can efficiently build virtual HPC clusters on the cloud; (ii) the new adaptive fault tolerance strategy outperforms other existing strategies based on blocking restoration; (iii) the integration of singularity containers into HPC@Cloud improves the portability of HPC applications to public clouds with negligible performance penalty to the applications; (iv) the proposed cost prediction approach can estimate the cost of running the applications on AWS and Vultr with up to 93% accuracy on average.

TEE-Graph: efficient privacy and ownership protection for cloud-based graph spectral analysis.

Big graphs like social network user interactions and customer rating matrices require significant computing resources to maintain. Data owners are now using public cloud resources for storage and computing elasticity. However, existing solutions do not fully address the privacy and ownership protection needs of the key involved parties: data contributors and the data owner who collects data from contributors. We propose a Trusted Execution Environment (TEE) based solution: TEE-Graph for graph spectral analysis of outsourced graphs in the cloud. TEEs are new CPU features that can enable much more efficient confidential computing solutions than traditional software-based cryptographic ones. Our approach has several unique contributions compared to existing confidential graph analysis approaches. (1) It utilizes the unique TEE properties to ensure contributors' new privacy needs, e.g., the right of revocation for shared data. (2) It implements efficient access-pattern protection with a differentially private data encoding method. And (3) it implements TEE-based special analysis algorithms: the Lanczos method and the Nystrom method for efficiently handling big graphs and protecting confidentiality from compromised cloud providers. The TEE-Graph approach is much more efficient than software crypto approaches and also immune to access-pattern-based attacks. Compared with the best-known software crypto approach for graph spectral analysis, PrivateGraph, we have seen that TEE-Graph has 103-105 times lower computation, storage, and communication costs. Furthermore, the proposed access-pattern protection method incurs only about 10%-25% of the overall computation cost. Our experimentation showed that TEE-Graph performs significantly better and has lower costs than typical software approaches. It also addresses the unique ownership and access-pattern issues that other TEE-related graph analytics approaches have not sufficiently studied. The proposed approach can be extended to other graph analytics problems with strong ownership and access-pattern protection.

Deadline-constrained cost-aware workflow scheduling in hybrid cloud

Cloud computing offers resources to users for computing and storage needs on the policy pay-what-you-use model. A private cloud is an excellent cost-saving option because the user owns the resources and can be used free to execute workflow applications. However, when private cloud resources are insufficient to fulfill the needs of workflow applications, the public cloud is the only option. To this end, a hybrid cloud (HC) is an elegant way to combine public and private cloud resources to meet the requirements of workflow applications. Therefore, large enterprises prefer the HC, but task scheduling in the HC is complicated. Therefore, this study proposed a Deadline-constrained Cost-aware Workflow Scheduling (DCWS) algorithm to address task scheduling problems in the HC. Firstly, we will execute maximum workflow tasks on the private cloud owned by users under the deadline constraint. Secondly, send the unscheduled workflow tasks to the public cloud for execution. Lastly, we will ensure scheduling whole workflow tasks on the private cloud and the communication channels (for sending to the public cloud) to satisfy the task-precedence needs and the task deadline constraint while incurring a minimal monetary cost. Furthermore, we reduce task slack time with the help of the sub-deadline division technique and search for an ideal virtual machine (VM) to reduce further execution time and monetary cost. The experimental results prove that the DCWS algorithm outperformed existing algorithms.

Neural simulation pipeline: Enabling container-based simulations on-premise and in public clouds.

In this study, we explore the simulation setup in computational neuroscience. We use GENESIS, a general purpose simulation engine for sub-cellular components and biochemical reactions, realistic neuron models, large neural networks, and system-level models. GENESIS supports developing and running computer simulations but leaves a gap for setting up today's larger and more complex models. The field of realistic models of brain networks has overgrown the simplicity of earliest models. The challenges include managing the complexity of software dependencies and various models, setting up model parameter values, storing the input parameters alongside the results, and providing execution statistics. Moreover, in the high performance computing (HPC) context, public cloud resources are becoming an alternative to the expensive on-premises clusters. We present Neural Simulation Pipeline (NSP), which facilitates the large-scale computer simulations and their deployment to multiple computing infrastructures using the infrastructure as the code (IaC) containerization approach. The authors demonstrate the effectiveness of NSP in a pattern recognition task programmed with GENESIS, through a custom-built visual system, called RetNet(8 × 5,1) that uses biologically plausible Hodgkin-Huxley spiking neurons. We evaluate the pipeline by performing 54 simulations executed on-premise, at the Hasso Plattner Institute's (HPI) Future Service-Oriented Computing (SOC) Lab, and through the Amazon Web Services (AWS), the biggest public cloud service provider in the world. We report on the non-containerized and containerized execution with Docker, as well as present the cost per simulation in AWS. The results show that our neural simulation pipeline can reduce entry barriers to neural simulations, making them more practical and cost-effective.

SLA-Based Scheduling of Spark Jobs in Hybrid Cloud Computing Environments

Big data frameworks such as Apache Spark is becoming prominent to perform large-scale data analytics jobs in various domains. However, due to limited resource availability, the local or on-premise computing resources are often not sufficient to run these jobs. Therefore, public cloud resources can be hired on a pay-per-use basis from the cloud service providers to deploy a Spark cluster entirely on the cloud. Nevertheless, using only cloud resources can be costly. Hence, both local and cloud resources nowadays are used together to deploy a hybrid cloud computing cluster. However, scheduling jobs in a cluster deployed on hybrid clouds is challenging in the presence of various Service-Level Agreement (SLA) demands such as cost minimization and job deadline guarantee. Most of the existing works either consider a public or a locally deployed cluster and mainly focus on improving job performance in the cluster. In this article, we propose efficient scheduling algorithms that leverage from different VM instance pricing in a hybrid cloud deployed cluster to optimize the Virtual Machine (VM) usage cost for both local and cloud resources and maximize the job deadline met percentage. We have conducted extensive simulation-based experiments to compare our proposed algorithms with the baseline approaches. In addition, we have developed a prototype system on top of Apache Mesos cluster manager and performed real experiments to evaluate the applicability of our proposed approaches in a real platform with benchmark applications. The results show that our proposed algorithms are highly scalable and reduce the cost of VM usage of a hybrid cluster for up to 20 percent.

Training and Serving ML workloads with Kubeflow at CERN

Machine Learning (ML) has been growing in popularity in multiple areas and groups at CERN, covering fast simulation, tracking, anomaly detection, among many others. We describe a new service available at CERN, based on Kubeflow and managing the full ML lifecycle: data preparation and interactive analysis, large scale distributed model training and model serving. We cover specific features available for hyper-parameter tuning and model metadata management, as well as infrastructure details to integrate accelerators and external resources. We also present results and a cost evaluation from scaling out a popular ML use case using public cloud resources, achieving close to linear scaling when using a large number of GPUs.

Application of binary PSO for public cloud resources allocation system of video on demand (VoD) services

Video streaming, whether on demand or live, has become one of the most popular internet applications. However, financial investments required for it is a severe problem since it needs more real time storage, higher data transfer and a significant amount of computation than other kinds of multimedia data. To tackle this problem, cloud computing, offering services without investing in hardware or software, emerges as a preferred technology. However, there are a large number of cloud service providers and they offer different pricing strategies for various applications in various regions. Therefore, it is of great importance for them that incoming service requests are assigned to the appropriate cloud services at minimum cost and provide maximum user satisfaction [quality of service (QoS) attributes]. Due to the issues, such as multiple cloud providers, different QoS requirements, different service level agreements and uncertainties in demand, price and availability, the optimization of resource allocation present further challenges. The objective of our study is to optimize the cost and performance of video on demand applications using cloud content delivery networks, storage and transcoders based on the QoS requirements of users. To solve the NP-hard problem, Particle Swarm Optimization (PSO) technique is used due to the easiness in its concept and coding, less sensitive to the nature of the objective function, limited number of parameters and generating high quality solution within a short time. We propose a new method in which the optimum solution is affected not only by the best solution of the particle and global best solution but also by the best solution of the neighborhood particles in that iteration. This ternary approach is implemented into the well-known discrete and constrained PSO, achieving the minimum cost with user satisfaction for allocation of video requests to cloud resources. Although the proposed method yields better results in terms of accuracy, execution time of the algorithm is not reasonable. To overcome this inefficiency; ternary approach is embedded into multi-swarm PSO and it is parallelized and combined with greedy heuristic algorithms. The results of the comparison with the benchmarking algorithms show that our proposed method yields better results from the standpoint of both accuracy and execution time.

Dynamic scheduling of bags-of-tasks with sensitive input data and end-to-end deadlines in a hybrid cloud

As organizations with existing on-premise infrastructure investments shift to the hybrid cloud computing paradigm, it is imperative to address the various challenges involved. One of the most important issues is the utilization of novel workload scheduling heuristics in order to effectively harness the security provided by the private cloud and the virtually unlimited resources of the public cloud. In this paper, we propose heuristics for the scheduling of real-time bag-of-tasks jobs that arrive dynamically at a hybrid cloud. The proposed scheduling strategies take into account the end-to-end deadlines of the jobs, as well as the monetary cost required for the utilization of the complementary public cloud resources. Furthermore, they take into consideration that some of the component tasks of the jobs may require input data that are sensitive and thus should not be transferred to the public cloud. The performance of the proposed heuristics is evaluated by simulation. For comparison purposes, two widely used baseline scheduling policies are also examined. In the simulation experiments, we consider jobs with either tight or loose deadlines and with different probabilities that the input data of their component tasks are sensitive.

Generative Adversarial Networks for fast simulation

Deep Learning techniques are being studied for different applications by the HEP community: in this talk, we discuss the case of detector simulation. The need for simulated events, expected in the future for LHC experiments and their High Luminosity upgrades, is increasing dramatically and requires new fast simulation solutions. Here we present updated results on the development of 3DGAN, one of the first examples using three-dimensional convolutional Generative Adversarial Networks to simulate high granularity electromagnetic calorimeters. In particular, we report on two main aspects: results on the simulation of a more general, realistic physics use case and on data parallel strategies to distribute the training process across multiple nodes on public cloud resources.

Shared data-aware dynamic resource provisioning and task scheduling for data intensive applications on hybrid clouds using Aneka

In the recent years, data-intensive applications have been growing at an increasing rate and there is a critical need to solve the high-performance and scalability issues. Hybrid Cloud Computing paradigm provides a promising solution to harness local infrastructure and remote resources and provide high Quality of Service (QoS) for time sensitive and data-intensive applications. Generally, hybrid cloud deployments have a heterogeneous pool of resources and it becomes a challenging task to efficiently utilize resources to provide optimum results. In modern data hungry applications, it is crucial to optimize bandwidth consumption, latency and networking overheads. Moreover, most of them have large extent of file sharing capability. The existing algorithms do not explicitly consider file sharing scenarios that leads large data transmission times and has severe effects on latency. In this direction, this paper focuses on building upon existing dynamic resource provisioning and task scheduling algorithms to provide better QoS in hybrid cloud environments for data intensive applications in a shared file task environment. The efficiency of proposed algorithms is demonstrated by deploying them on Microsoft Azure using Aneka, a platform for developing scalable applications on the Cloud. Experiments using real-world applications and datasets show that proposed algorithms are able to allocate tasks and extend to public cloud resources more efficiently, reducing deadline violations and improving response times to give response time reduction of upto 40.12% for a sample local alignment search application on genome sequences.

A Cost Estimation Model for Cloud Services and Applying to PC Laboratory Platforms

IaaS (Infrastructure as a Service) is a well-known computing service, which provides infrastructures over the cloud without owning real hardware resources. This is attractive as resources can be scaled up and down instantly according to the user’s computing demands. Customers of such services would like to adjust the utilization policy promptly by considering the charge of the service, but an instantaneous response is not possible as it takes several hours or even a couple of days for cloud service providers to inform the billing information. In this article, we present an instant cost estimation model for estimating the cost of public cloud resources. Specifically, our model estimates the cost of IaaS by monitoring the usage of resources on behalf of virtual machine instances. As this is performed by generating a user-side metering daemon, it is very precise and thus similar to the resource usage evaluated by the cloud service provider. To validate our model, we run PC laboratory services for 50 students in two classes by making use of a public cloud during a semester. Experimental results show that the accuracy of our model is over 99.3% in comparison with the actual charge of the public cloud.

A Decentralized Accountability Framework For Enhancing Secure Data Sharing Through Icm In Cloud

The Cloud substitutes a computing criterion where shared configurable resources are afforded as an on-demand service over the Internet. Moreover, the cloud environment provides resources to the users on the basis of services like SaaS, PaaS and IaaS. Generally, a cloud can be referred as private cloud or public cloud. When a Cloud Service Provider (CSP) imposes upon public cloud resources to compile their private cloud, the result is demonstrated as a virtual private cloud. Private or public, the imperious intent of cloud computing is to provide simplistic, reliable usage of various computing resources. One of the significant features of cloud is that the outsourced data are accessed through any anonymous machines over the Internet. On the other hand, it creates an issue that user’s fear of unknown access of data, which can become a major difficulty to the wide implementation of cloud. In this paper, a decentralized accountability framework is developed to monitor the actual usage and access of the data that is shared on cloud. For that, a logging mechanism that includes authentication for each user to access the data has also been provided. Moreover, some procedures for providing the data under the control of data owner includes Integrity Checking Mechanism (ICM) have also been developed. The overall process strengthens the security constraints over cloud. And the experimental results reveal that the approach affords secure and scalable data sharing with reduced memory utilization and processing time.

A Decentralized Accountability Framework for Enhancing Secure Data Sharing Through ICM in Cloud

The Cloud substitutes a computing criterion where shared configurable resources are afforded as an on-demand service over the Internet. Moreover, the cloud environment provides resources to the users on the basis of services like SaaS, PaaS and IaaS. Generally, a cloud can be referred as private cloud or public cloud. When a Cloud Service Provider (CSP) imposes upon public cloud resources to compile their private cloud, the result is demonstrated as a virtual private cloud. Private or public, the imperious intent of cloud computing is to provide simplistic, reliable usage of various computing resources. One of the significant features of cloud is that the outsourced data are accessed through any anonymous machines over the Internet. On the other hand, it creates an issue that user’s fear of unknown access of data, which can become a major difficulty to the wide implementation of cloud. In this paper, a decentralized accountability framework is developed to monitor the actual usage and access of the data that is shared on cloud. For that, a logging mechanism that includes authentication for each user to access the data has also been provided. Moreover, some procedures for providing the data under the control of data owner includes Integrity Checking Mechanism (ICM) have also been developed. The overall process strengthens the security constraints over cloud. And the experimental results reveal that the approach affords secure and scalable data sharing with reduced memory utilization and processing time.

A fault‐tolerant dynamic scheduling method on hierarchical mobile edge cloud computing

AbstractMobile edge cloud computing has been a promising computing paradigm, where mobile users could offload their application workloads to low‐latency local edge cloud resources. However, compared with remote public cloud resources, conventional local edge cloud resources are limited in computation capacity, especially when serve large number of mobile applications. To deal with this problem, we present a hierarchical edge cloud architecture to integrate the local edge clouds and public clouds so as to improve the performance and scalability of scheduling problem for mobile applications. Besides, to achieve a trade‐off between the cost and system delay, a fault‐tolerant dynamic resource scheduling method is proposed to address the scheduling problem in mobile edge cloud computing. The optimization problem could be formulated to minimize the application cost with the user‐defined deadline satisfied. Specifically, firstly, a game‐theoretic scheduling mechanism is adopted for resource provisioning and scheduling for multiprovider mobile applications. Then, a mobility‐aware dynamic scheduling strategy is presented to update the scheduling with the consideration of mobility of mobile users. Moreover, a failure recovery mechanism is proposed to deal with the uncertainties during the execution of mobile applications. Finally, experiments are designed and conducted to validate the effectiveness of our proposal. The experimental results show that our method could achieve a trade‐off between the cost and system delay.

PrivateGraph: Privacy-Preserving Spectral Analysis of Encrypted Graphs in the Cloud

Big graphs, such as user interactions in social networks and customer rating matrices in collaborative filters, possess great values for both businesses and research. They are not only big but often keep evolving, which requires a large amount of computing resources to maintain. With the wide deployment of public cloud resources, owners of big graphs may want to use cloud resources to obtain storage and computation scalability. However, privacy and ownership of the graphs in the cloud has become a major concern. In this paper, we study privacy-preserving algorithms for one of the important graph analysis techniques—graph spectral analysis for outsourced graph in the cloud. The core operation: eigendecomposition of large matrix is also important to many data mining algorithms. We consider a cloud-centric framework with three collaborative parties: data contributors, data owner, and cloud provider. Graphs are represented as matrices such as adjacency matrix and Laplacian matrix, the elements of which are encrypted and submitted by distributed contributors. The data owner then interacts with the cloud-side programs to conduct spectral analysis, while protecting data privacy from the honest-but-curious cloud provider. For a $N\times N$N×N graph matrix, we aim to design algorithms with the cloud handling expensive storage and computation in $O(N^2)$O(N2) complexity, while data owner and data contributors’ algorithms take only $O(N)$O(N). To achieve this goal, we develop the privacy-preserving versions of the two approximate eigendecomposition algorithms: the Lanczos algorithm and the Nystrom algorithm, considering two encryption methods: additive homomorphic encryption (AHE) methods and somewhat homomorphic encryption (SHE) methods. Both dense and sparse matrices are studied, while sparse matrices also involve a differentially private data submission protocol to allow the trade-off between data sparsity and privacy. Experimental results show that the Nystro algorithm with sparse encoding can dramatically reduce data owners’ costs. SHE-based methods have lower computational time while AHE-based methods have lower communication/storage costs.

Efficient deadline-aware scheduling for the analysis of Big Data streams in public Cloud

The emergence of Big Data has had a profound impact on how data are analyzed. Open source distributed stream processing platforms have gained popularity for analyzing streaming Big Data as they provide low latency required for streaming Big Data applications using Cloud resources. However, existing resource schedulers are still lacking the efficiency and deadline meeting that Big Data analytical applications require. Recent works have already considered streaming Big Data characteristics to improve the efficiency and the likelihood of deadline meeting for scheduling in the platforms. Nevertheless, they have not taken into account the specific attributes of analytical application, public Cloud utilization cost and delays caused by performance degradation of leasing public Cloud resources. This study, therefore, presents BCframework, an efficient deadline-aware scheduling framework used by streaming Big Data analysis applications based on public Cloud resources. BCframework proposes a scheduling model which considers public Cloud utilization cost, performance variation, deadline meeting and latency reduction requirements of streaming Big Data analytical applications. Furthermore, it introduces two operator scheduling algorithms based on both a novel partitioning algorithm and an operator replication method. BCframework is highly adaptable to the fluctuation of streaming Big Data and the performance degradation of public Cloud resources. Experiments with the benchmark and real-world queries show that BCframework can significantly reduce the latency and utilization cost and also minimize deadline violations and provisioned virtual machine instances.