Search

Cloud computing enables clients to acquire cloud resources dynamically and on demand for their cloud applications and services. For cloud providers, especially, Software as a Service (SaaS) providers, the prediction of future cloud resource requirements, such as CPU usage for their cloud applications, to implement client requests is a complex task because it depends on incoming workloads. Due to workload fluctuations, it is difficult for SaaS cloud providers to predict or forecast future demand for resource usage in the next time interval and, accordingly, to allocate the required resources. Furthermore, cloud computing systems consist of many virtual machines (VMs), which increases the complexity of the prediction problem due to the correlations that exist between the large workload data in these VMs. Therefore, accurate resource usage forecasting remains a challenge, and relatively few studies have explored the prediction of CPU usage for VMs in cloud data centers. This paper proposes an autonomic and intelligent workload forecasting method for cloud resource provisioning based on the concept of autonomic computing and a deep learning approach. In particular, to predict future demand for CPU usage and determine how to respond to workload fluctuations in the next interval, we propose an efficient deep learning model based on a diffusion convolutional recurrent neural network (DCRNN). Existing deep learning models that are widely applied cannot handle accurate real-time forecasting due to the presence of inconsistent and nonlinear workloads in cloud computing systems. The goal of the proposed deep learning model is to improve forecasting accuracy and minimize the error between the predicted and the actual workloads. The effectiveness of the proposed DCRNN-based deep learning model was evaluated using experiments on a real-world dataset of PlanetLab’s CPU usage traces. The results indicate that the proposed approach outperformed other existing deep learning models, achieving a mean absolute percentage error of 0.18 and root-mean-square error of 2.40.

This paper presents a Power and Resource Utilization-Aware Virtual Machine Scheduling (PRUVMS) algorithm for strengthening resource utilization and diminishing the energy consumption of servers in the cloud environment. The PRUVMS algorithm enhances the resource utilization by migrating the VMs from the underloaded/overloaded servers to a normal server, and it reduces the energy consumption by shutting down the underloaded servers after migrating the VMs. For selecting the suitable server for the VM placement, the ranking of the available servers is evaluated. An illustrative example is presented to validate the PRUVMS algorithm. Further, the PRUVMS algorithm is tested on the PlanetLab workload using the CloudSim simulator. The proposed PRUVMS algorithm improves resource utilization by 68.22% and 37.53% and decreases the energy consumption by 35.53% and 31.34% in comparison with PABFD and CAVMP algorithms, respectively. The improvement in computational results shows the acceptability of the proposed scheduling algorithm in the cloud environment.

Cloud service providers use load balancing algorithms in order to avoid Service Level Agreement Violations (SLAVs) and wasted energy consumption due to host over- and under-utilization, respectively. Load balancing algorithms migrate VMs between hosts in order to balance host loads. Any Virtual Machines (VMs) that are migrated experience performance degradation which results in lower Quality of Service (QoS) and can possibly result in SLAVs. Hence, an optimal load balancing method should reduce the number of over- and under-utilized hosts with a minimal number of VM migrations. One of the metrics used previously in the literature for evaluating load balancing stated that it equally considered SLAVs caused by both over-utilized hosts and migrations. However, in this paper, we show that, in fact, this metric values keeping the number of migrations low at the expense of an increased number of over-utilized hosts. This disparity is demonstrated by simulation of Google, PlanetLab and Azure data sets in CloudSim. This metric may suit public cloud providers which are focused on minimizing SLAVs and keeping energy costs low, but does not consider the QoS of customer VMs. We propose an alternative metric that considers QoS for the VMs. This alternative metric considers not only performance loss during migration, but also performance degradation due to host over-utilization. Private cloud providers, e.g., IT services within large organizations, often value the performance of their “customer” VMs, i.e., the QoS their organization receives, as well as traditional cloud provider costs, i.e., energy and SLAV costs. Hence, our alternative metric would be more appropriate in these scenarios. We compare and contrast load balancing methods using both the existing, biased metric and our new alternative metric.

Network caching in named data networks (NDN) is essential for improving the potentialities of the conventional IP networking. The concept of network caching is necessary for achieving optimal bandwidth utilization and location independent data access during multipath data dissemination. However, network caching in NDN makes it highly vulnerable to security breaches such as access content packets violation, flooding or malicious injection of packets and content cache poisoning. In this paper, a dynamic forest of random subsets-based one-time signature-based capability enhancing security architecture (DFORS-CSA) is proposed for attaining distributed data authentication. This DFORS-CSA security architecture leverages the potential in exploring the access privileges of the packets disseminated in the network. It includes the capability through which the routes can perform authentication of packets forwarded in NDN. It supports a significant verification strategy through which the routers can ensure the packet timeliness for resolving the problems that get introduced through unsolicited packets exchanged during flooding-based denial of service attacks. The simulation experiments of the proposed DFORS-CSA is conducted using the open source CCNs platform and Planetlab simulator. The results of the proposed DFORS-CSA confirmed its predominance in minimizing overall delay and time incurred in the bit vector generation by 16.74 and 15.63%, excellent to the baseline approaches. The results of the proposed DFORS-CSA also conformed a mean improvement in the precision rate by 10.21%, true positive rate by 8.94% and F-measure by 7.62% with decreased false positive rate of 8.56%, during the process of detecting content cache poisoning attack.

The cloud computing environments rely heavily on virtualization that enables the physical hardware resources to be shared among cloud users by creating virtual machines (VMs). With an overloaded physical machine, the resource requests by virtual machines may not be fulfilled, which results in Service Level Agreement (SLA) violations. Moreover, the high performance servers in cloud data centers consume large amount of energy. The dynamic VM consolidation techniques use live migration of virtual machines to optimize resource utilization and minimize energy consumption. An excessive migration of virtual machines may however deteriorate application performance due to the overhead incurring at runtime. In this paper, we propose a normalization-based VM consolidation (NVMC) strategy that aims at placing virtual machines in an online manner while minimizing energy consumption, SLA violations, and the number of VM migrations. The proposed strategy uses resource parameters for determining over-utilized hosts in a virtualized cloud environment. The comparative capacity of virtual machines and hosts is incorporated for determining over-utilized hosts, while the cumulative available-to-total ratio (CATR) is used to find under-utilized hosts. For migrating virtual machines to appropriate hosts, the VM placement uses a criteria based on normalized resource parameters of hosts and virtual machines. For evaluating the performance of VM consolidation, we have performed experimentation with a large number of virtual machines using traces from the PlanetLab workloads. The results show that the NVMC approach outperforms other well-known approaches by achieving a significant improvement in energy consumption, SLA violations, and number of VM migrations.

Cloud Data Centers (CDCs) have become a vital computing infrastructure for enterprises. However, CDCs consume substantial energy due to the increased demand for computing power, especially for the Internet of Things (IoT) applications. Although a great deal of research in green resource allocation algorithms have been proposed to reduce the energy consumption of the CDCs, existing approaches mostly focus on minimizing the number of active Physical Machines (PMs) and rarely address the issue of load fluctuation and energy efficiency of the Virtual Machine (VM) provisions jointly. Moreover, existing approaches lack mechanisms to consider and redirect the incoming traffics to appropriate resources to optimize the Quality of Services (QoSs) provided by the CDCs. We propose a novel adaptive energy-aware VM allocation and deployment mechanism called AFED-EF for IoT applications to handle these problems. The proposed algorithm can efficiently handle the fluctuation of load and has good performance during the VM allocation and placement. We carried out extensive experimental analysis using a real-world workload based on more than a thousand PlanetLab VMs. The experimental results illustrate that AFED-EF outperforms other energy-aware algorithms in energy consumption, Service Level Agreements (SLA) violation, and energy efficiency.

Proliferation of large number of cloud users steered the exponential increase in number and size of the data centers. These data centers are energy hungry and put burden for cloud service provider in terms of electricity bills. There is environmental concern too, due to large carbon foot print. A lot of work has been done on reducing the energy requirement of data centers using optimal use of CPUs. Virtualization has been used as the core technology for optimal use of computing resources using VM migration. However, networking devices also contribute significantly to the responsible for the energy dissipation. We have proposed a two level energy optimization method for the data center to reduce energy consumption by keeping SLA. VM migration has been performed for optimal use of physical machines as well as switches used to connect physical machines in data center. Results of experiments conducted in CloudSim on PlanetLab data confirm superiority of the proposed method over existing methods using only single level optimization.

Workflows are the de facto standard for managing business processes and allow businesses to automate interactions between business locations and partners residing anywhere on the planet. This, however, requires the workflows to be executed in a dynamic environment, where device and communication failures occur frequently, making availability a key concern. In this work, we propose the replicated execution of workflows for ensuring availability in the presence of failures. The replicated execution has to yield the same result as a non-replicated execution of that workflow. Thus, we formally define Single-Execution-Equivalence and present a replication scheme that adheres to this definition. We implement a proof-of-concept using an open-source workflow engine for demonstrating the compatibility with current workflow technology. Our evaluations on Amazon EC2, OpenStack, and PlanetLab show that workflow replication ensures availability while being scalable and incurring low overhead in terms of execution time.

Traditional data centers are shifted toward the cloud computing paradigm. These data centers support the increasing demand for computational and data storage that consumes a massive amount of energy at a huge cost to the cloud service provider and the environment. Considerable energy is wasted to constantly operate idle virtual machines (VMs) on hosts during periods of low load. Dynamic consolidation of VMs from overloaded or underloaded hosts is an effective strategy for improving energy consumption and resource utilization in cloud data centers. The dynamic consolidation of VM from an overloaded host directly influences the service level agreements (SLAs), utilization of resources, and quality of service (QoS) delivered by the system. We proposed an algorithm, namely, GradCent, based on the Stochastic Gradient Descent technique. This algorithm is used to develop an upper CPU utilization threshold for detecting overloaded hosts by using a real CPU workload. Moreover, we proposed a dynamic VM selection algorithm called Minimum Size Utilization (MSU) for selecting the VMs from an overloaded host for VM consolidation. GradCent and MSU maintain the trade-off between energy consumption minimization and QoS maximization under specified SLA goal. We used the CloudSim simulations with real-world workload traces from more than a thousand PlanetLab VMs. The proposed algorithms minimized energy consumption and SLA violation by 23% and 27.5% on average, compared with baseline schemes, respectively.

Real time resource usage prediction is an important part of resource provisioning in a cloud data centre. As cloud workloads vary dynamically, effective resource provisioning requires prediction of future resource usage trends. The problem is highly complicated because of highly time varying nature of cloud resource workloads. Training the future resource usage prediction models once, using a fixed set of observations is not sufficient to capture the variability in cloud workloads. In this work, we propose to use gradient descent (GD) and Levenberg-Marquardt (LM) adaptation algorithms for dynamic adaptation of resource utilization prediction models. We also propose a novel sparse framework for fast online adaptation of resource usage prediction models. We propose to analyze different algorithms such as $\ell ^{1}$ regularization, $\ell ^{2}$ regularization, optimal brain damage (OBD), optimal brain surgeon (OBS) for introducing sparsity. The proposed sparse framework for online adaptation of multivariate resource usage prediction models is validated for CPU usage prediction in the Google cluster trace and PlanetLab workload trace. A comparative analysis of different sparse frameworks shows that OBD-based LM adaptation algorithm performs better than other frameworks for online multivariate resource usage prediction in a cloud.

Energy-efficient Cloud Infrastructure Resource Allocation Framework is getting popularity as it is paying effective attention to cloud data management with a view to achieve maximize revenue and minimize cost. This infrastructure can encourage for both cloud providers and users for allocating cloud infrastructure resources for fulfilling not only good energy efficiency measured in Power Usage Effectiveness (PUE) and data center Infrastructure Efficiency (DCiE) but also high CPU utilization. Therefore, in this paper we proposed a framework which can show effective performance for achieving the high data center energy efficiency and preventing Service Level Agreement (SLA) violation respectively with the aim of green cloud resources deployment. The framework accomplishes cloud infrastructure resource allocation on the basic of Reinforcement Learning mechanism and Fuzzy Logic for green solutions. The evaluation for Energy-efficient Resource Allocation is experimented on the traces of the PlanetLab virtualized environment for gaining good PUE and CPU utilization.

In recent years, cloud data centers are rapidly growing with a large number of finite heterogeneous resources to meet the ever-growing user demands with respect to the SLA (service level agreement). However, the potential growth in the number of large-scale data centers leads to large amounts of energy consumption, which is constantly a major challenge. In addition to this challenge, intensive number of VM (virtual machine) migrations can decrease the performance of cloud data centers. Thus, how to minimize energy consumption while satisfying SLA and minimizing the number of VM migrations becomes an important challenge classified as NP-hard optimization problem in data centers. Most VM scheduling schemes have been proposed for this problem, such as dynamic VM consolidation. However, most of them failed in low time complexity and optimal solution. Hence, this paper proposes a dynamic VM consolidation approach-based load balancing to minimize the trade-off between energy consumption, SLA violations and VM migrations while keeping minimum host shutdowns and low time complexity in heterogeneous environment. Specifically, the proposed approach consists of four methods which include: BPSO meta-heuristic-based load balancing to impact on energy consumption and number of host shutdowns, overloading host detection and VM placement-based Pearson correlation coefficient to impact on SLA, and VM selection based on imbalance degree to impact on number of VM migration. Moreover, Pearson correlation coefficient and imbalance degree correlate CPU, RAM and bandwidth, respectively, in each host and each VM. Through extensive analysis and simulation experiments using real PlanetLab and random workloads, the performance results demonstrate that the proposed approach exhibits excellent results for the NP-problem.

The cloud service providers require a large number of computing resources to provide services on-demand that consume the electricity at large and leave high carbon footprints which must be minimized. A cloud system must optimally use its resources to achieve a low operational cost without degrading the quality of services. In this context, an ensemble learning based workload forecasting method is presented that uses extreme learning machines and their corresponding forecasts are weighted by a voting engine. A metaheuristic algorithm inspired by blackhole theory is used to select the optimal weights. The accuracy of the approach is tested on CPU and memory demand requests of Google cluster trace. The method is also compared with recent existing work in the literature on CPU utilization of Google cluster and PlanetLab traces. The results validate the superiority of the approach over existing methods with an improvement up to 99.20% in root mean squared error.

Virtualization provides the flexibility to distribute the workload among physical servers to reduce overall electrical power consumption, through the consolidation of Virtual Machines (VMs). Many research projects have been done on VM migration to reduce energy consumption in data centers while ensuring a high level of adherence to the Service Level Agreements (SLA). Service levels of running applications are likely to be negatively affected during a live VM migration. For this reason, in this paper, we propose a new intelligent VM migration approach, called CLANFIC, which utilizes modified Cellular Learning Automata based Evolutionary Computing (CLA-EC) and neuro-fuzzy to minimize the number of VM migrations and improve energy consumption. This goal is achieved by utilizing an optimized placement method and delaying migration time based on future resource demand prediction. This algorithm reduces the number of migrations in two steps (i) finding the optimal virtual machine placement and replacement on physical servers by using modified CLA-EC (ii) predicting future resource usage in each host by a neuro-fuzzy algorithm to prevent unnecessary migrations. The experimental results on the real workload traces from PlanetLab show that the proposed method reduces the mean migration number, energy consumption, and SLA violation of the data center by 59.05%, 8.5%, and 70.76%, respectively.

Deep penetration of personal computing devices and high-speed Internet has enabled everyone to be a broadcaster. In this crowdsourced live streaming service, numerous amateur broadcasters lively stream their video contents to viewers around the world. Consequently, these broadcasters generate a massive amount of video data. The set of video sources and recipients are big as well, so are demand for the storage and computational resources. Transcoding becomes a must to better service these viewers with different network and device configurations. However, the massive amount of video data contributed by countless channels even makes cloud significantly expensive for providing transcoding services to the whole community. In this paper, inspired by the paradigm of Edge Computing, we propose a Cloud-edge collaborative system which combines the idle end-viewers’ resources with the cloud to transcode the massive amount of videos at scale. Specifically, we put forward tailored viewer selection algorithms after empirically analyses the viewer behavior data. In the meantime, we propose auction-based payment schemes to motivate these viewers participating in the transcoding. Large-scale trace-driven simulations demonstrate the superiority of our approach in cost reduction and service stability. We further implement a prototype in PlanetLab to prove the feasibility of our design.

Rapid data streams are generated continuously from diverse sources including users, devices, and sensors located around the globe. This results in the need for efficient geo-distributed streaming analytics to extract timely information. A typical geo-distributed analytics service uses a hub-and-spoke model, comprising multiple edges connected by a wide-area-network (WAN) to a central data warehouse. In this paper, we focus on the widely used primitive of windowed grouped aggregation , and examine the question of how much computation should be performed at the edges versus the center . We develop algorithms to optimize two key metrics: WAN traffic and staleness (delay in getting results). We present a family of optimal offline algorithms that jointly minimize these metrics, and we use these to guide our design of practical online algorithms based on the insight that windowed grouped aggregation can be modeled as a caching problem where the cache size varies over time. We evaluate our algorithms through an implementation in Apache Storm deployed on PlanetLab. Using workloads derived from anonymized traces of a popular analytics service from a large commercial CDN, our experiments show that our online algorithms achieve near-optimal traffic and staleness for a variety of system configurations, stream arrival rates, and queries.

Dynamic consolidation of virtual machines (VMs) is presented as a significant technique of energy conservation in cloud environments, which can eliminate the hotspot of overloaded hosts and switch the under loaded hosts to sleep mode through live migration of virtual machines. However, since the fact that migrating VM consumes a certain amount of extra resources, the process of reallocation can cause Service Level Agreement (SLA) violations. In this paper, a novel proactive framework which considers both predicted resource utilization and Performance-to-power Ratio (PPR) of heterogeneous hosts is proposed to perform dynamic VM consolidation to achieve balance of performance and energy. More precisely, a workload predictor is proposed based on the modified Weighted Moving Average (WMA) algorithm, representing the support for dynamic resource allocation; a cluster controller is proposed based on reinforcement learning for exploring the optimal matching relationship between resource requests and host at different PPR levels; a resource allocator is designed based on greedy strategy for achieving the trade-off between energy consumption and application performance across the cluster. Moreover, the framework is implemented based on distributed architecture and off-line learning pattern, which are able to not only scale up quickly but also improve the computing efficiency of the system. To validate the effectiveness of the proposed method, we have performed experimental evaluation on CloudSim with real-world workload traces of PlanetLab, and the simulation results demonstrate that it reduces the energy consumption up to 45.25% and effectively deals with high Quality of Service (QoS) requirements and heterogeneous distributed infrastructures in comparison with other competitive approaches.

Studies on the resource workload demand in cloud computing environment aim at reducing resource wastage by optimizing the resource utilization in a cloud data center. Based on this goal, most of the existing approaches rely on resource management mechanisms such as resource allocation and Virtual Machine (VM) consolidation to reach an ideal solution for reducing resource wastage. Because of instability and high variability of the cloud resource usage and workloads, there is a demand for cloud providers to apply the prediction methods for forecasting the future cloud resource utilization. This paper employs a supervised statistical learning method, i.e., Support Vector Regression Technique (SVRT), to forecast the future usage of multi-attribute host resource. The method is particularly suitable to handle a non-linear cloud resource workload. To improve the prediction accuracy of SVRT, we decide Radial Basis Function as the kernel function of SVRT and apply Sequential Minimal Optimization Algorithm (SMOA) for the training and regression estimation of the prediction method. Besides, compared with the existing work, we consider the multi-attribute cloud resources other than the single resource. The method is employed under eight sets of real-world workloads, which are collected from BitBrain (BB), PlanetLab (PL) and Google Cluster Workload Traces (GCWT). Series of experiments conducted on the workload dataset show the effectiveness of our approach. Based on evaluation metrics, the final results show that the accuracy was enhanced by approximately 4%-16% and the error percentage was reduced by approximately 8%-60% compared with the state-of-the-art methods.

In cloud data centers, an appropriate Virtual Machine (VM) placement has become an effective method to improve the resource utilization and reduce the energy consumption. However, most current solutions regard the VM placement as a bin-packing problem and each VM is seen as a single object. None of them have taken the relationships between VMs into consideration, which supplies us with a kind of new perspective. In this paper, we provide a model which explores the relationships for every two VMs based on the resource requirement provided by ARIMA prediction. This model evaluates the volatility of resource utilization after putting the two VMs on the same host and we call this model as affinity model. Based on the affinity model, VMs will be placed on those hosts that have the highest affinity with them. Therefore, we call it as Predicted Affinity based Virtual Machine Placement Algorithm (PAVMP). The advantages of PAVMP are showed by comparing it with other VM placement algorithms on CloudSim simulation platform with the PlanetLab and Google workload trace.

Network programmability has metamorphosed over the past twenty years from the controversial research vision of active networks, through PlanetLab, to the juggernaut of SDN and OpenFlow that has swept industry. Now PISA switches are emerging with support for protocol-independent reconfigurability. We reflect on how network architecture has evolved along a different path than we had foreseen to arrive at a place that is not so different than we and other researchers had hoped and imagined.

The virtual machine (VM) workload of a datacenter is dynamic, where the reallocation of a subset of active VMs can result in better VM allocation by avoiding over-loaded/under-loaded physical machines (PMs). Over-loaded PMs lead to customer dissatisfaction, whereas under-loaded PMs result in increased energy consumption. In this work, we propose a multi-objective best-fit-decreasing (BFD) approach to the VM reallocation problem. Our multi-objective formulation considers power costs and resource utilization. We use the expressive power of fuzzy algebra to combine both objectives into a single-objective function. Extensive simulations, using CloudSim, show that our fuzzy-based multi-objective implementation of BFD leads to significantly better solutions with respect to energy and resource utilization. Indeed, the results show an improvement of as much as 30% to 40% of energy consumption and 30% of resource utilization when compared with reported heuristics which minimize energy only, using five real workloads provided as a part of the coMon project, which is a monitoring infrastructure for PlanetLab.

Cloud providers leverage live migration of virtual machines to reduce energy consumption and allocate resources efficiently in data centers. Each migration decision depends on three questions: when to move a virtual machine, which virtual machine to move and where to move it? Dynamic, uncertain, and heterogeneous workloads running on virtual machines make such decisions difficult. Knowledge-based and heuristics-based algorithms are commonly used to tackle this problem. Knowledge-based algorithms, such as MaxWeight scheduling algorithms, are dependent on the specifics and the dynamics of the targeted Cloud architectures and applications. Heuristics-based algorithms, such as MMT algorithms, suffer from high variance and poor convergence because of their greedy approach. We propose an online reinforcement learning algorithm called Megh. Megh does not require prior knowledge of the workload rather learns the dynamics of workloads as-it-goes. Megh models the problem of energy- and performance-efficient resource management during live migration as a Markov decision process and solves it using a functional approximation scheme. While several reinforcement learning algorithms are proposed to solve this problem, these algorithms remain confined to the academic realm as they face the curse of dimensionality. They are either not scalable in real-time, as it is the case of MadVM, or need an elaborate offline training, as it is the case of Q-learning. These algorithms often incur execution overheads which are comparable with the migration time of a VM. Megh overcomes these deficiencies. Megh uses a novel dimensionality reduction scheme to project the combinatorially explosive state-action space to a polynomial dimensional space with a sparse basis. Megh has the capacity to learn uncertain dynamics and the ability to work in real-time without incurring significant execution overhead. Megh is both scalable and robust. We implement Megh using the CloudSim toolkit and empirically evaluate its performance with the PlanetLab and the Google Cluster workloads. Experiments validate that Megh is more cost-effective, converges faster, incurs smaller execution overhead and is more scalable than MadVM and MMT. An empirical sensitivity analysis explicates the choice of parameters in experiments.

Hybrid peer-to-peer assisted cloud-based video-on-demand (VoD) systems augment cloud-based VoD systems with P2P networks to improve scalability and save bandwidth costs in the cloud. In these systems, the VoD service provider (e.g., NetFlix) relies on the cloud to deliver videos to users and pays for the cloud bandwidth consumption. The users can download videos from both the cloud and peers in the P2P network. It is important for the VoD service provider to i) minimize the cloud bandwidth consumption, and ii) guarantee users’ satisfaction (i.e., quality-of-experience). Though previous adaptive bitrate streaming (ABR) methods improve video playback smoothness, they cannot achieve these two goals simultaneously. To tackle this challenge, we propose AutoTune, a game-based adaptive bitrate streaming method. In AutoTune, we formulate the bitrate adaptation problem in ABR as a noncooperative Stackelberg game, where VoD service provider and the users are players. The VoD service provider acts as a leader and it decides the VoD service price for users with the objective of minimizing cloud bandwidth consumption while ensuring users’ participation. In response to the VoD service price, the users select video bitrates that lead to maximum utility (defined as a function of its satisfaction minus associated VoD service fee). Finally, the Stackelberg equilibrium is reached in which the cloud bandwidth consumption is minimized while users are satisfied with selected video bitrates. To enhance the performance of AutoTune, we further propose the reputation-based incentive scheme and the popularity-based cache management scheme. Experimental results from the PeerSim simulator and the PlanetLab real-world testbed show that compared to existing methods, AutoTune can provide high user satisfaction and save cloud bandwidth consumption. Also, the proposed enhancement schemes are effective in improving the performance of AutoTune.