Yunlan Wang

Optimizing the fault-tolerance overheads of HPC systems using prediction and multiple proactive actions

The complexity and scale of high-performance computer systems are rapidly increasing, so fault tolerance is becoming a critical challenge. In this paper, we consider the impact of multiple proactive actions on proactive fault tolerance and periodic checkpointing. We extended Aupy’s model in the presence of multiple proactive actions, including proactive checkpointing and task migration. We then propose optimal strategies for deciding when to trust predictions, and provide algorithms for the optimal storage interval for periodic checkpointing. The results show that the proposed method can significantly improve system productivity. Our case study indicates that the recall of the predictor is more important for small platforms, and that precision becomes increasingly important as the scale of the system increases.

Two-Level Trust Federation Model for Cross-Domain Environment

Under distributed environment, Resources are located in different security domains. The cross-domain network application pattern is formed to access cross-domain resource, which is adapted to industry application and become widely-used network scheme. Therefore, it is necessary to build a loose and scalable dynamic trust federation, which requires the establishment of trust relationship between various entities. Through a study of trust mechanism, this paper divides entity into two parts including inner-domain and inter-domain and presents a two-level trust federation model in a cross-domain environment. And based on this model, this paper realizes the trust establishment and trust management for different security domains. Meanwhile, this paper describes the problem of privacy protection, information security, access control and federation evaluation in the model. The dynamic trust federation model achieves the trust interaction among different entities. It can expand the trust function of SAML components in real application while realizes the two-level trust federation model, which is called trust-based SAML framework (TBSF). At the same time, TBSF achieve cross-domain access in industrial collaborative design and application.

Campus Computational Grid oriented Resource Optimizing Management

Resource management is one of the crucial issues in the campus grid environment. In this paper, we propose a resources optimizing management scheme in the campus computational grid environment. The NPU Campus computational grid is introduced. Then, we propose some resources optimizing management approaches. The application features are presented and analyzed

An Incremental and Hash-based Algorithm for Mining Frequent Episodes

Episodes rules can describe and predict the behavior of the event sequences. The property of incremental frequent episodes mining is studied and the related lemmas and corollaries are presented, then a general incremental algorithm named IHE for mining frequent episodes is proposed. Moreover, it proposes and utilizes the window-hash-based technique to prune candidate episodes. The performance of the algorithm IHE was evaluated and compared with the algorithm WINEPI. It is shown by our experimental results that the algorithm IHE has better performance

Research on Optimum Checkpoint Interval for Hybrid Fault Tolerance

With the rapid growth of the high performance computer system size and complexity, passive fault tolerance can no longer effectively provide reliability of the system because of the high overhead and poor scalability of these methods. Hybrid fault tolerant method which is the combination of passive and active fault tolerant approaches has the potential to be widely used in fault tolerance of exascale system. However, there are still many issues of this method need to be ironed out. This paper focuses on the issues of checkpointing of hybrid fault tolerant method. A common question surrounding checkpointing is the optimization of the checkpoint interval. This paper proposes two models to model the systems which adopt hybrid fault tolerance. By comparing their results with the simulation, this paper evaluates the effectiveness of these two models. Experimental result shows that the modified model can not only predict the total work time excellently, but also can predict the optimum checkpoint interval precisely.

mHLogGP: A Parallel Computation Model for CPU/GPU Heterogeneous Computing Cluster

CPU/GPU heterogeneous computing has become a tendency in scientific and engineering computing. The conventional computation models cannot be used to estimate the application running time under the CPU/GPU heterogeneous computing environment. In this paper, a new model named mHLogGP is presented on the basis of mPlogP, LogGP and LogP. In mHLogGP, he communication and memory access is abstracted based on the characteristic of CPU/GPU hybrid computing cluster. This model can be used to study the behavior of application, estimate the execution time and guide the optimization of parallel programs. The results show that the predicted running time approaches to the actual execution of program.

Prediction of Blasting Vibration Intensity by Improved PSO-SVR on Apache Spark Cluster

In order to predict blasting vibration intensity accurately, support vector machine regression (SVR) was adopted to predict blasting vibration velocity, vibration frequency and vibration duration. The mutation operation of genetic algorithm (GA) is used to avoid the local optimal solution of particle swarm optimization (PSO). The improved PSO algorithm is used to search for the best parameters of SVR model. In the experiments, the improved PSO-SVR algorithm was realized on the Apache Spark platform. The execution time and prediction accuracy of the sadovski method, the traditional SVR algorithm, the neural network (NN) algorithm and the improved PSO-SVR algorithm were compared. The results show that the improved PSO-SVR algorithm on Spark is feasible and efficient, and the SVR model can predict the blasting vibration intensity more accurately than other methods.

Managing high-performance computing applications as an on-demand service on federated clouds

Abstract There are several challenges (e.g., imbalance between supply and demand of hardware resources and software licenses, and usability) under modern High-Performance Computing (HPC) environment. As a means of providing an on-demand service for end users, we propose a Software-as-a-Service (SaaS) approach for managing commercial HPC applications as a Web-based service deployed on top of federated clouds. Some inter-trusted private or public clouds are federated to create a unified service platform with a large amount of hardware resources. In addition, an on-demand, pay-per-use model for Web-service-enabled HPC applications is proposed. Further, we provide an economic analysis of the proposed approach from the perspective of end users, cloud service providers, and Independent Software Vendors (ISVs). We conduct a simulation using two HPC application services on three federated clouds. A combined Quality of Service (QoS) and economic evaluation demonstrates a better effect of the proposed approach comparing with existing HPC platforms.

A new design and implementation of scientific workflow simulator

In this paper, a novel scientific workflow simulator is designed and implemented which is energy-aware and can simulate the scientific workflow that running on the multi-core cluster system. A method for application modeling is proposed which can describe the process dependence, data dependence and performance requirement of the workflow. A computing system model was also introduced to describe the layered structure of the cluster, the communication matrixes of the cluster nodes, and the energy consumption under different load level. Based on the application model and the computing systems model, the scientific workflow scheduling problem was abstracted to multi-objective optimization problem, a scheduling algorithm is presented which can satisfy the performance constraints and is energy aware as while. The experiment results proved the effectiveness of the simulator.

An autonomic monitoring framework of web service-enabled application software for the hybrid distributed HPC infrastructure

For the distributed high performance computing infrastructure, there is a service oriented trend with hybrid physical clusters and virtual clusters. Application software can be service-enabled on the basis of underlying hardware resources. To enable on-demand service for end users, monitoring of software services is important in the service oriented hybrid computing environment. In this paper, an autonomic monitoring framework for the web service-enabled application software is proposed. Software resources are web service-enabled on the basis of elastic hardware resources—physical machines or virtual machines. Both of the software services and underlying hardware resources are dynamically monitored. An autonomic monitoring algorithm with self-optimized updating period and self-adaptive event-driven update is also given for the dynamic information retrieving of software services and underlying hardware resources. Some experiments were conducted in a multi-cluster based hybrid distributed HPC infrastructure. The proposed approach can bring more accurate monitoring information with a better updating cost.