Weidong Bao

FESTAL: Fault-Tolerant Elastic Scheduling Algorithm for Real-Time Tasks in Virtualized Clouds

As clouds have been deployed widely in various fields, the reliability and availability of clouds become the major concern of cloud service providers and users. Thereby, fault tolerance in clouds receives a great deal of attention in both industry and academia, especially for real-time applications due to their safety critical nature. Large amounts of researches have been conducted to realize fault tolerance in distributed systems, among which fault-tolerant scheduling plays a significant role. However, few researches on the fault-tolerant scheduling study the virtualization and the elasticity, two key features of clouds, sufficiently. To address this issue, this paper presents a fault-tolerant mechanism which extends the primary-backup model to incorporate the features of clouds. Meanwhile, for the first time, we propose an elastic resource provisioning mechanism in the fault-tolerant context to improve the resource utilization. On the basis of the fault-tolerant mechanism and the elastic resource provisioning mechanism, we design novel fault-tolerant elastic scheduling algorithms for real-time tasks in clouds named FESTAL, aiming at achieving both fault tolerance and high resource utilization in clouds. Extensive experiments injecting with random synthetic workloads as well as the workload from the latest version of the Google cloud tracelogs are conducted by CloudSim to compare FESTAL with three baseline algorithms, i.e., Non-M igration-FESTAL (NMFESTAL), Non-Overlapping-FESTAL (NOFESTAL), and Elastic First Fit (EFF). The experimental results demonstrate that FESTAL is able to effectively enhance the performance of virtualized clouds.

SP-Partitioner: A novel partition method to handle intermediate data skew in spark streaming

Abstract Spark Streaming, a popular tool for processing live data streams, offers a good divide-and-conquer solution, where data stream is split into batches that are then processed in parallel by mappers, and the intermediate data from the mappers are finally reduced by reducers. However, one of the key issues with such an approach for live data processing is partitioning skew in which data distributed over the processing units are not balanced due to uncertainty of the coming data streams. This imbalance is rippled through the mappers and become prominent to the reducers, making reduce a performance bottleneck to the overall system. To address this issue, we present a Partitioner, SP-Partitioner, that sits between the map and reduce stages to re-balance the workload of the reducers. With our design, we treat the arrived batches of data as candidate samples and choose samples based on systematic sampling to predict the characteristics of intermediate data. According to the prediction, our method generates a reference table to guide the allocation of next batches of data evenly. We implement SP-Partitioner in Spark 1.6.1 and evaluate its performance with widely used applications. Experimental results conducted on a real VMs cluster show that our algorithms can not only achieve higher balancing performance on data with varying degree of data skew, but also decrease the average processing time of one batch of these data.

Improving the Performance of Data Sharing in Dynamic Peer-to-Peer Mobile Cloud

Mobile cloud computing has become an emerging computing paradigm to extend the capability of the mobile devices and it has gained increasing popularity in recent years. Existing studies mainly focus on how to leverage the computing capability of the individual device by employing the capability from remote cloud datacenters or local mobile cloud formed by nearby devices. Different from these studies, we investigate how to improve the performance of data sharing in the peer-to-peer mobile cloud, with the limited bandwidth and the presence of dynamic and unpredictable wireless channel state. Specifically, we first formulate the data transmission among devices as a utility maximization problem with the consideration of limited bandwidth, incentive participation and the QoE (Quality of Experience) heterogeneity, based on incorporating publish/subscribe component into the base station. Then, a dynamic online algorithm, which does not need the future context (e.g., channel state) of the mobile cloud, is developed to simultaneously make the decision of data transmission and communication interface selection. Rigorously theoretical analysis shows the optimality and the effectiveness of the proposed algorithm. Extensive experiments are conducted to verify the analysis results and the superiority of the proposed algorithm over existing strategies.

Towards collaborative storage scheduling using alternating direction method of multipliers for mobile edge cloud

We propose a collaborative storage architecture of mobile edge cloud.We propose a collaborative storage scheduling algorithm named ACMES.ACMES minimizes power usage and withdrawal risk with assured reliability.ACMES works in a distributed and parallel way.We conduct extensive experiments to validate the superiority of ACMES. Performance of cloud computing would be much improved by extending storage capabilities to devices at the edge of network. Unfortunately, the commonly employed algorithms fail to be adaptive to the new storage pattern on mobile edge cloud. To address this issue, we propose a collaborative storage architecture model and an alternating-direction-method-of-multipliers-based collaborative storage scheduling algorithm called ACMES (Algorithm of Collaborative Mobile Edge Storage), in which heterogeneous information of nodes in mobile edge cloud is considered and integrated to make decisions. Besides, feasible solutions for storage will be acquired after iterations of computing. By formulating the collaborative storage scheduling problem in the mobile edge cloud and designing the collaborative decision-making process with the theory of Alternating Direction Method of Multipliers (ADMM), the proposed ACMES is able to minimize power usage and the risk of node withdrawal without reducing the reliability of node storage, and meanwhile make storage scheduling decisions at the edge environment directly and work in a distributed and parallel way. The convergence analysis shows that ACMES has the ability to solve complicated mobile edge cloud storage problems in reality. Extensive experiments validate its effectiveness as well as its superiority to three existing strategies (ADM, RDM and ERASURE) in total cost, reliability, power usage and withdrawal risks.

MECCAS: Collaborative Storage Algorithm Based on Alternating Direction Method of Multipliers on Mobile Edge Cloud

The commonly existing employed centralized algorithms fail to be adaptive to the new storage pattern on mobile edge cloud. To address this issue, we propose an alternating-direction-method-of-multipliers-based collaborative storage algorithm called MECCAS (Mobile Edge Cloud Collaborative Storage). The proposed MECCAS is able to minimize the delay of task execution and total costs for the overall operation, and meanwhile maximize the utilization of local information of nodes and system reliability. Nodes on mobile edge cloud storage are capable of adaptively allocating resources for storage to increase power usage effectiveness and reduce the risk of nodes withdrawal. Extensive experiments demonstrate the superiority of our MECCAS algorithm compared with other three baselines, i.e., ADM, RDM and ERASURE. The optimization utility of our algorithm is higher than other three algorithms by 41.72%, 44.52% and 22.94% on average, respectively

CHIME: A Checkpoint-Based Approach to Improving the Performance of Shared Clusters

Due to the limitation of resources, preemption frequently occurs in almost all the commercial cloud platforms, such as Google cluster and Amazon cluster. Since preemption can ensure that once the system is in heavy workload, high-priority tasks will be executed primarily and at the same time, some low-priority tasks will be killed immediately. Then when more resources are available, the killed tasks will restart to execute. Especially, during the peak time, some low-priority tasks could possibly be preempted and restarted repeatedly resulting in much more consuming precious resources including CPU cores, RAM and hard drives. Thanks to the checkpoint technology, it provides an efficient solution to addressing the preemption issue. But checkpoint technology has limitations, e.g., making checkpoint frequently will add redundant overhead to the cluster and cause I/O congestion. In this paper, by leveraging checkpoint technology, we designed a novel approach to improving the performance of shared clusters. Specifically, by checking the occupancy of resources periodically, making decisions to checkpoint or not and checkpointing for certain tasks, our method can reduce unnecessary checkpoints and exalt the performance of the whole cloud, especially tasks with low-priority. Extensive simulation experiments injecting tasks following the Google cloud trace logs were conducted to validate the superiority of our approach by comparing it with some baselines.

DEFT: Dynamic Fault-Tolerant Elastic scheduling for tasks with uncertain runtime in cloud

Abstract With the widespread use of clouds, the reliability and efficiency of cloud have been the main concerns of the service providers and users. Thus, fault tolerance has become a hotspot in both industry and academia, especially for real-time applications. To achieve fault tolerance in cloud, a great number of in-depth researches have been conducted. Nevertheless, for addressing the issue of fault tolerance, few studies have taken into account the uncertainty of task runtime, which is however more practical and really needs urgent attention. In this paper, we introduce the uncertainty to our task runtime estimation model and we propose a fault-tolerant task allocation mechanism that strategically uses two fault tolerant task scheduling models while the uncertainty is considered. Moreover, we employ the overlapping mechanism to improve the resource utilization of cloud. Based on the two mechanisms, we propose an innovative D ynamic F ault- T olerant E lastic scheduling algorithm-DEFT for scheduling real-time tasks in the cloud where the system performance volatility should be considered. The purpose of DEFT is to achieve both fault tolerance and resource utilization efficiency. We compare DEFT with three baseline algorithms: NDRFT, DRFT, and NWDEFT. The results from our extensive experiments on the workload of the Google tracelogs show that DEFT can guarantee fault tolerance while achieving high resource utilization.

WITCAT: A Workload Spike Targeted Cloud Management Solution.

The cloud computing technology offers consistent access to large-scale computing capabilities, thereby bringing convenience to life. However, the virtualized cloud systems are still too vulnerable to maintain performance scalability and service agility once a task burst surges in without any warning. A mounting account of research has been conducted on proper strategies for accurate workload prediction as well as effective resource reservation and arrangement, but commonly cloud providers seek help to strategies that deploy excessive resources, adding overhead cost and sacrificing the clouds advantage of scalability, or otherwise fail to reconfigure timely and properly, causing dissatisfaction and even financial loss, which are not expected by both cloud providers and clients. In this paper, we present a holistic solution called Workload Spike Targeted Cloud Management Solution (WITCAT) for virtualized cloud systems with three fundamental modules as a whole, which was seldom proposed before. By learning historical taskflow patterns, WITCAT can effectively classify the arriving tasks into clusters that feature respective workload traits. Then two different prediction means are employed to continually forecast the arrival rate and attributes of workloads for respective clusters, under two different characteristic scenarios: normal scenarios and bursty scenarios. Last, we employ a reservation strategy, makes full use of the available resources, strengthening the effectiveness of cloud service provisioning under workload spike. As far as our knowledge reaches, the contributions are three-fold. • We improve the clustering method for task characterization, where a Mahalanobis-distance-bused k-means clustering is adopted to eliminate the relevance among tasks attributes. • We employ a traffic-oriented two-scenario integrated prediction method, with a control knob that monitors the increment of tasks and triggers prediction means alternation for different workload scenarios. • We develop a prediction-based heuristic algorithm for resource reservation and provisioning, reserving enough space in CPU and memory ahead of time for bursts without disabling the clouds scalibility. We conduct extensive experiments using Google cloud traces and the results outperform other scheduling algorithms in guarantee ratio (25.8% improved), total energy consumption (17.3% saved) and resource utilization (18.2% improved), which further indicates the advantages of our proposed solution towards task traffic bursts.

Not Just Privacy: Improving Performance of Private Deep Learning in Mobile Cloud

The increasing demand for on-device deep learning services calls for a highly efficient manner to deploy deep neural networks (DNNs) on mobile devices with limited capacity. The cloud-based solution is a promising approach to enabling deep learning applications on mobile devices where the large portions of a DNN are offloaded to the cloud. However, revealing data to the cloud leads to potential privacy risk. To benefit from the cloud data center without the privacy risk, we design, evaluate, and implement a cloud-based framework ARDEN which partitions the DNN across mobile devices and cloud data centers. A simple data transformation is performed on the mobile device, while the resource-hungry training and the complex inference rely on the cloud data center. To protect the sensitive information, a lightweight privacy-preserving mechanism consisting of arbitrary data nullification and random noise addition is introduced, which provides strong privacy guarantee. A rigorous privacy budget analysis is given. Nonetheless, the private perturbation to the original data inevitably has a negative impact on the performance of further inference on the cloud side. To mitigate this influence, we propose a noisy training method to enhance the cloud-side network robustness to perturbed data. Through the sophisticated design, ARDEN can not only preserve privacy but also improve the inference performance. To validate the proposed ARDEN, a series of experiments based on three image datasets and a real mobile application are conducted. The experimental results demonstrate the effectiveness of ARDEN. Finally, we implement ARDEN on a demo system to verify its practicality.

A General Cross-Layer Cloud Scheduling Framework for Multiple IoT Computer Tasks

The diversity of IoT services and applications brings enormous challenges to improving the performance of multiple computer tasks’ scheduling in cross-layer cloud computing systems. Unfortunately, the commonly-employed frameworks fail to adapt to the new patterns on the cross-layer cloud. To solve this issue, we design a new computer task scheduling framework for multiple IoT services in cross-layer cloud computing systems. Specifically, we first analyze the features of the cross-layer cloud and computer tasks. Then, we design the scheduling framework based on the analysis and present detailed models to illustrate the procedures of using the framework. With the proposed framework, the IoT services deployed in cross-layer cloud computing systems can dynamically select suitable algorithms and use resources more effectively to finish computer tasks with different objectives. Finally, the algorithms are given based on the framework, and extensive experiments are also given to validate its effectiveness, as well as its superiority.