Amazon EC2 Research Articles

Multi-Objective Workflow Scheduling With Deep-Q-Network-Based Multi-Agent Reinforcement Learning

Cloud Computing provides an effective platform for executing large-scale and complex workflow applications with a pay-as-you-go model. Nevertheless, various challenges, especially its optimal scheduling for multiple conflicting objectives, are yet to be addressed properly. The existing multi-objective workflow scheduling approaches are still limited in many ways, e.g., encoding is restricted by prior experts' knowledge when handling a dynamic real-time problem, which strongly influences the performance of scheduling. In this paper, we apply a deep-Q-network model in a multi-agent reinforcement learning setting to guide the scheduling of multi-workflows over infrastructure-as-a-service clouds. To optimize multi-workflow completion time and user's cost, we consider a Markov game model, which takes the number of workflow applications and heterogeneous virtual machines as state input and the maximum completion time and cost as rewards. The game model is capable of seeking for correlated equilibrium between make-span and cost criteria without prior experts' knowledge and converges to the correlated equilibrium policy in a dynamic real-time environment. To validate our proposed approach, we conduct extensive case studies based on multiple well-known scientific workflow templates and Amazon EC2 cloud. The experimental results clearly suggest that our proposed approach outperforms traditional ones, e.g., non-dominated sorting genetic algorithm-II, multi-objective particle swarm optimization, and game-theoretic-based greedy algorithms, in terms of optimality of scheduling plans generated.

Scheming an Proficient Auto Scaling Technique for Minimizing Response Time in Load Balancing on Amazon AWS Cloud

Cloud computing is one of the developing innovation in the field of data improvement. Because of its huge application, for example, virtualization and dynamic asset portion, the cloud computing ends up being the most encouraging innovation. To give the solid support of the customers with Quality of Service (QOS), the load balancing component with auto scaling methods would be vital in cloud computing conditions. With the end goal to keep the framework from overloading and slamming, a productive calculation must be given by the application and approaching client activity in the cloud computing condition. The proposed Auto Load Distribution (ALD) calculation have been assessed and contrasted and some of existing load balancing algorithm and its effectiveness is vastly improved than the past calculations. To accomplish the previously mentioned issues, we tried our undertakings in Amazon EC2 stage, one of open cloud administrations which give correct qualities. Results have been investigated and limiting the reaction time greater by utilization of auto scaling system through our proposed ALD algorithm.

A low-latency computing framework for time-evolving graphs

The demand to deliver fast responses in processing time-evolving graphs is higher than ever before in a large number of big data applications. This problem promotes extensive uses of an incremental computing model, which executes the underlying graph algorithm on the newly updated graph structure by taking the results of the computation on the outdated graph structure as initial values, in distributed time-evolving graph computing systems. In this paper, we experimentally study how the initial values of the computation on a newly updated graph structure influence the convergence of the iterative graph analysis, and we develop an optimization framework on the basis of the incremental computing model to accelerate the convergence of processing time-evolving graphs thus achieving high performance for time-evolving graph analysis. In contrast to the traditional incremental computing model, which uses the results of the computation on the outdated graph structure directly, the proposed framework predicts the optimal initial values of the computation on the new graph structure and thereby reduces the number of iterations. Two different prediction approaches are designed to optimize the initial values based on a combination of the results of the computation on the previous graph data and the newly incoming graph data. We have evaluated our optimization framework using the graph algorithms PageRank and KMeans on Amazon EC2 clusters. The experiments demonstrate that the incremental computing implementation with the initial value prediction have reduced the number of iterations by 30% for the PageRank algorithm and 13.7% for the KMeans algorithm, and reduced the response time by 12.7% and 10.6% accordingly compared to the traditional incremental computing model.

Cost-Effective Algorithm for Workflow Scheduling in Cloud Computing Under Deadline Constraint

Cloud computing is a popular model that allows users to store, access, process, and retrieve data remotely. It provides a high-performance computing with large scale of resources. However, this model requires an efficient scheduling strategy for resources management. Recently, several algorithms are presented to solve the resource scheduling problem. Nevertheless, still the problem exists with complex applications such as workflows, which need an efficient algorithm to be scheduled on the available resources. This paper presents a novel hybrid algorithm, called CR-AC, combining both the chemical reaction optimization (CRO) and ant colony optimization (ACO) algorithms to solve the workflow-scheduling problem. The proposed CR-AC algorithm is implemented in the CloudSim toolkit and evaluated by using real applications and Amazon EC2 pricing model. Moreover, the results are compared with the most recent algorithms: modified particle swarm optimization (PSO) and cost-effective genetic algorithm (CEGA). The experimental results indicate that the CR-AC algorithm achieves better results than the traditional CRO, the ACO, the modified PSO and CEGA algorithms, in terms of total cost, time complexity, and schedule length.

A Game Theoretic Approach to Estimate Fair Cost of VM Placement in Cloud Data Center

Pricing of virtual machines (VMs) with different dimensions is a challenging task. VM pricing involves both capital and operational expenditures. Capital expenditure is fixed in nature, while operational expenditure is variable. A fraction of capital expenditure is included for VM pricing. An individual pays the cloud service provider (CSP) for his requested VM. If the users cooperate among themselves they may end up paying less to CSP for their requested VMs vis-a-vis they would have paid had they requested individually. In this paper, an n -person cooperative game is adopted to determine the price that users would pay for their requested VMs under a cooperative environment. Shapley value is used to estimate the fraction of capital expenditure that would be included in the VM price. An integer linear programming is proposed for energy-efficient placement. For evaluation, VM configurations and pricing of popular CSPs—Microsoft Azure and Amazon EC2—are considered. Results show that users would pay less for their requested VMs if they cooperate. The energy consumed by the proposed VM placement technique is compared with first fit decreasing (FFD) and enhanced first fit decreasing (EFFD). It is observed that the proposed technique consumes lesser energy compared to FFD and EFFD.

Low-Complexity Distributed Radio Access Network Slicing: Algorithms and Experimental Results

Radio access network (RAN) slicing is an effective methodology to dynamically allocate networking resources in 5G networks. One of the main challenges of RAN slicing is that it is provably an NP-Hard problem. For this reason, we design near-optimal low-complexity distributed RAN slicing algorithms. First, we model the slicing problem as a congestion game, and demonstrate that such game admits a unique Nash equilibrium (NE). Then, we evaluate the Price of Anarchy (PoA) of the NE, i.e., the efficiency of the NE as compared with the social optimum, and demonstrate that the PoA is upper-bounded by 3/2. Next, we propose two fully-distributed algorithms that provably converge to the unique NE without revealing privacy-sensitive parameters from the slice tenants. Moreover, we introduce an adaptive pricing mechanism of the wireless resources to improve the network owner’s profit. We evaluate the performance of our algorithms through simulations and an experimental testbed deployed on the Amazon EC2 cloud, both based on a real-world dataset of base stations from the OpenCellID project. Results conclude that our algorithms converge to the NE rapidly and achieve near-optimal performance, while our pricing mechanism effectively improves the profit of the network owner.

ECloud - Magnetotelluric Webapp

Geophysical web applications are highly underrated. There are many potential benefits for transitioning to a cloud based web applications including compatibility improvement, low on-going operating costs and better access to existing third-party code and tools for facilitating rapid development. For users, time spent on installation, general IT maintenance such license management and upgrading is minimized. While web applications for business style applications are deeply established, scientific applications on the web are only just emerging. This research builds upon our previous feasibility study (Pethick and Harris 2015), which was the development of a 1D MT web application. This is to be redevelop into a commercial grade cloud hosted geophysical inversion web app that can cooperatively invert seismic and magnetotelluric (MT) data. Our software, ECloud, is designed to be user friendly where geoscientists and drillers can upload MT field data directly from their laptop, tablet or mobile to obtain subsurface geo-electrical distributions quickly with minimal input. This application is designed to be scalable, suiting cloud environments and is currently hosted on an Amazon EC2 instance. The preconceived notion that web applications are slow will be challenged. The purposefully designed MT algorithm and software structure will hopefully result in lower computation times while minimizing restrictions based on hardware requirements (i.e., primary memory).

Multiagent system for dynamic resource provisioning in cloud computing platforms

Dynamic provisioning of computational resources in cloud platforms is a challenging problem. In this scenario, multiagent system technology may offer noticeable improvements, with intelligent agents to dynamically choose the best computational load distribution reducing execution time and cloud services cost. Thus, this article presents MAS-Cloud, a multiagent system to dynamically monitor, predict and provide computational resources in cloud platforms. Deductive reasoning agents work cooperatively in a three layer architecture to provide transparent horizontal elasticity of virtual machines in public cloud platforms (i.e., Google, Amazon EC2). MAS-Cloud was evaluated with a nondeterministic and CPU intensive application providing a challenging validation case. A multiple linear regression model was used presenting good predictions with reduced mean absolute percent error achieving 3.59% CPU usage and 5.28% of execution time in the Google platform, and 3.97% CPU usage and 5.47% of execution time in the Amazon EC2 platform.

Truthful Online Auction Toward Maximized Instance Utilization in the Cloud

Although infrastructure as a service (IaaS) users are busy scaling up/out their cloud instances to meet the ever-increasing demands, the dynamics of their demands, as well as the coarse-grained billing options offered by leading cloud providers, have led to substantial instance underutilization in both temporal and spatial domains. This paper systematically examines an instance subletting service, where sublettable instances can be leased to others within predetermined periods when underutilized, from both theoretical and practical perspectives. The studied instance subletting service extends and complements the existing instance market of IaaS providers. We identify the unique challenges and opportunities in this new service, and design online auction mechanisms to make allocation and pricing decisions for the instances to be sublet. For static supplies of instances, our mechanism guarantees truthfulness and individual rationality with the best possible competitive ratio. We then incorporate a multi-stage discount strategy to gracefully handle dynamic supplies. Extensive trace-driven simulations show that our service achieves significant performance gains in both cost savings and social welfare. We further pinpoint the challenges in implementing such a service in the real-world system and validate our modeling assumptions through a container-based prototype implemented over Amazon EC2.

Dynamic Load Balancing Model Based on Server Status (DLBS) for Green Computing

There could be problems with server quality when resources are limited and this could lead to poor service delivery to the clients. Therefore, a strong load balancing technique is required that helps in the optimization of resource utilization. Cloud servers with correlated load balancer can assist in optimizing the load balancing practicability in cloud computing. In this paper, a load balancing model (DLBS) is proposed for effective load balancing and resource optimization in public cloud. A load balancing model was designed and implemented using Amazon EC2. Each server was equipped with a load balancer which monitors the load and sends status information to the controller. The servers with fewer loads were given more requests while the overloaded ones were not given further requests. The Amazon Web Services (AWS) was used to demonstrate the proof of concept. The results revealed that the proposed solution improved performance, throughput, and utilization of cloud resources.

Generalized Numerical Entanglement for Reliable Linear, Sesquilinear and Bijective Operations on Integer Data Streams

We propose a new technique for the mitigation of fail-stop failures and/or silent data corruptions (SDCs) within linear, sesquilinear or bijective (LSB) operations on $M$ integer data streams ( $M\geq 3$ ). In the proposed approach, the $M$ input streams are linearly superimposed to form $M$ numerically entangled integer data streams that are stored in-place of the original inputs, i.e., no additional (aka. “checksum”) streams are used. An arbitrary number of LSB operations can then be performed in $M$ processing cores using these entangled data streams. The output results can be extracted from any $M-K$ entangled output streams by additions and arithmetic shifts, thereby mitigating $K$ fail-stop failures ( $K\leq \left\lfloor \frac{M-1}{2}\right\rfloor$ ), or detecting up to $K$ SDCs per $M$ -tuple of outputs at corresponding in-stream locations. Therefore, unlike other methods, the number of operations required for the entanglement, extraction and recovery of the results is linearly related to the number of the inputs and does not depend on the complexity of the performed LSB operations. Our proposal is validated within an Amazon EC2 instance (Haswell architecture with AVX2 support) via integer matrix product operations. Our analysis and experiments for fail-stop failure mitigation and SDC detection reveal that the proposed approach incurs 0.75 to 37.23 percent reduction in processing throughput in comparison to the equivalent error-intolerant processing. This overhead is found to be up to two orders of magnitude smaller than that of the equivalent checksum-based method, with increased gains offered as the complexity of the performed LSB operations is increasing. Therefore, our proposal can be used in distributed systems, unreliable multicore clusters and safety-critical applications, where robustness against failures and SDCs is a necessity.

A MOBILE CROWDSOURCING SYSTEM FOR FOOTBALL MATCH LIVE VIDEO STREAMING

Mobile crowdsourcing is a fast-growing emerging approach whereby large groups of mobile users are engaged in a collaborative work on performing a particular task or using its results. This paper presents a concept for the development of a mobile crowdsourcing system with extended capabilities for real-time broadcasting and receiving amateur football match video. It is designed to resolve the problem of possible delays and the overload of a system and to accelerate the process of big video data transmission. The proposed system is based on a service-oriented, three-layer cloud architecture and a specialized mobile video streaming application. The architecture includes a main server, infrastructure of scalable multi-parallel video processing engine and an auxiliary server for synchronizing real-time information, which significantly facilitates the handling of user requests with minimal cost and at a high speed. The concept is realized in the Footlikers platform as a basic client-server, WOWZA streaming engine, deployed on an Amazon EC2 cloud machine and a simple sync-server. The results of the program realization of the developed system prototype are presented, regarding football game video steaming intended for amateur football competitions based and organized in France, Belgium and Luxembourg.

Online Auction for IaaS Clouds: Towards Elastic User Demands and Weighted Heterogeneous VMs

Auctions have been adopted by many major cloud providers, such as Amazon EC2. Unfortunately, only simple auctions have been implemented. Such simple auction has serious limitations, such as being unable to accept elastic user demands and having to allocate different types of VMs independently. These limitations create a big gap between the real needs of cloud users and the available services of cloud providers. In response to the limitations of the existing auction mechanisms, this paper proposes a novel online auction mechanism for IaaS clouds, with the unique features of an elastic model for inputting time-varying user demands and a unified model for requesting heterogeneous VMs together. However, several major challenges should be addressed, such as NP hardness of optimal VM allocation, time-varying user demands and potential misreports of private information of cloud users. We propose a truthful online auction mechanism for maximizing the profit of the cloud provider in IaaS clouds, which is composed of a price-based allocation rule and a payment rule. In the allocation rule, the online auction mechanism determines the number of VMs of each type to each user. In the payment rule, by introducing a marginal price function for each type of VMs, the mechanism determines how much the cloud provider should charge each cloud user. With solid theoretical analysis and trace-driven simulations, we demonstrate that our mechanism is truthful, fair and individually rational, and has a polynomial-time complexity. In addition, our auction achieves a competitive ratio for the profit of the cloud provider, compared against the offline optimal one.

Analysis of Bounds on Hybrid Vector Clocks

Hybrid vector clock(s) (HVC) provide a mechanism to combine the theory and practice of distributed systems. Improving on traditional vector clock(s) (VC), HVC utilizes synchronized physical clocks to reduce the size by focusing only on causality where the physical time associated with two events is within a given uncertainty window $\epsilon$ and letting physical clock alone determine the order of events that are outside the uncertainty window. In this paper, we develop a model for determining the bounds on the size of HVC. Our model uses four parameters, $\epsilon$ : uncertainty window, $\delta$ : message delay, $\alpha$ : communication frequency and $n$ : number of nodes in the system. We derive the size of HVC in terms of a delay differential equation, and show that the size predicted by our model is almost identical to the results obtained by simulation. We also identify closed form solutions that provide tight lower and upper bounds for useful special cases. We show that for many practical applications and deployment environments in Amazon EC2, the size of HVC remains only as a couple entries and substantially less than $n$ . Finally, although the analytical results rely on a specific communication pattern they are useful in evaluating size of HVC in different communication scenarios.

Deploying Docker Swarm cluster on hybrid clouds using Occopus

The paper introduces a new method of deploying complex infrastructures in hybrid clouds. The new method is based on a new cloud orchestrator, called Occopus that enables cloud independent infrastructure deployment at user level. Occopus applies a plugin based cloud access architecture. As a result, porting infrastructure among different kind of clouds or deploying an infrastructure on top of a hybrid cloud including private and public (commercial) clouds becomes straightforward. Occopus supports all the major cloud technologies and protocols like OpenStack, OpenNebula, OCCI, Amazon EC2 or CloudSigma. The paper also shows how to apply Occopus in order to deploy a Docker Swarm cluster on the top of hybrid clouds. As a concrete example it shows the deployment on a hybrid cloud system containing OpenStack, Amazon and CloudSigma clouds as components. Using hybrid clouds raises security and communication performance issues that are also covered and investigated in the paper. The technology described in the paper is intensively used in the EU H2020 project COLA for more than 20 commercial applications and on MTA Cloud that is the cloud of the Hungarian Academy of Sciences where more than 40 academic projects run.

Cluster Programming using the OpenMP Accelerator Model

Computation offloading is a programming model in which program fragments (e.g., hot loops) are annotated so that their execution is performed in dedicated hardware or accelerator devices. Although offloading has been extensively used to move computation to GPUs, through directive-based annotation standards like OpenMP, offloading computation to very large computer clusters can become a complex and cumbersome task. It typically requires mixing programming models (e.g., OpenMP and MPI) and languages (e.g., C/C++ and Scala), dealing with various access control mechanisms from different cloud providers (e.g., AWS and Azure), and integrating all this into a single application. This article introduces computer cluster nodes as simple OpenMP offloading devices that can be used either from a local computer or from the cluster head-node. It proposes a methodology that transforms OpenMP directives to Spark runtime calls with fully integrated communication management, in a way that a cluster appears to the programmer as yet another accelerator device. Experiments using LLVM 3.8, OpenMP 4.5 on well known cloud infrastructures (Microsoft Azure and Amazon EC2) show the viability of the proposed approach, enable a thorough analysis of its performance, and make a comparison with an MPI implementation. The results show that although data transfers can impose overheads, cloud offloading from a local machine can still achieve promising speedups for larger granularity: up to 115× in 256 cores for the2MMbenchmark using 1GB sparse matrices. In addition, the parallel implementation of a complex and relevant scientific application reveals a 80× speedup on a 320 core machine when executed directly from the headnode of the cluster.

Evidence-Aware Mobile Computational Offloading

Computational offloading can improve user experience of mobile apps through improved responsiveness and reduced energy footprint. A fundamental challenge in offloading is to distinguish situations where offloading is beneficial from those where it is counterproductive. Currently, offloading decisions are predominantly based on profiling performed on individual devices . While significant gains have been shown in benchmarks, these gains rarely translate to real-world use due to the complexity of contexts and parameters that affect offloading. We contribute by proposing crowdsensed evidence traces as a novel mechanism for improving the performance of offloading systems. Instead of limiting to profiling individual devices, crowdsensing enables characterizing execution contexts across a community of users, providing better generalisation and coverage of contexts. We demonstrate the feasibility of using crowdsensing to characterize offloading contexts through an analysis of two crowdsensing datasets. Motivated by our results, we present the design and development of the EMCO toolkit and platform as a novel solution for computational offloading. Experiments carried out on a testbed deployment in Amazon EC2 Ireland demonstrate that EMCO can consistently accelerate app execution while at the same time reduce energy footprint. We also demonstrate that EMCO provides better scalability than current cloud platforms, being able to serve a larger number of clients without variations in performance. Our framework, use cases, and tools are available as open source from GitHub.

Power and resource-aware virtual machine placement for IaaS cloud

In the virtualization technology, cloud computing is a pool of abundant computing resources and delivers on-demand Internet-based computing services. One of the challenging issues in the virtualization is the placement of virtual machines (VMs) on the physical machines (PMs) such that the computing resources can be utilized efficiently. Furthermore, imbalanced usage of multi-dimensional resources may lead to overall resource wastage and SLA violations of a cloud data center. In this paper, we propose a new VM placement algorithm called multi-objective virtual machine placement (MOVMP) for IaaS cloud. In the algorithm, we devise a resource usage factor (RUF) to maximize the resource usage of the PMs during placement of the VMs. Further, we also present a multi-dimensional resource usage model, which direct to minimize the number of under-loaded PMs in IaaS cloud. This model also helps to improve resource utilization in a balanced manner and migrate a less number of VMs, which results in minimizing the resource wastage, power consumption, and the service level agreement (SLA) violations of cloud data center. The algorithm is tested using Amazon EC2 Instances. Through comparison results, we show that the proposed algorithm performs better than the existing ones in terms of various performance metrics.

Development of Real Time System for File Compression on Amazon EC2

Development of Real Time System for File Compression on Amazon EC2

An MCDM method for cloud service selection using a Markov chain and the best-worst method

Due to the increasing number of cloud services, service selection has become a challenging decision for many organisations. It is even more complicated when cloud users change their preferences based on the requirements and the level of satisfaction of the experienced service. The purpose of this paper is to overcome this drawback and develop a cloud broker architecture for cloud service selection by finding a pattern of the changing priorities of User Preferences (UPs). To do that, a Markov chain is employed to find the pattern. The pattern is then connected to the Quality of Service (QoS) for the available services. A recently proposed Multi Criteria Decision Making (MCDM) method, Best Worst Method (BWM), is used to rank the services. We show that the method outperforms the Analytic Hierarchy Process (AHP). The proposed methodology provides a prioritized list of the services based on the pattern of changing UPs. The methodology is validated through a case study using real QoS performance data of Amazon Elastic Compute (Amazon EC2) cloud services.