Ligu Zhu

An Improved Dynamic Load-Balancing Model

With the rapid development of big data and the national big data strategy is put forward, the Web server cluster facing more complex and severe challenges. The traditional load balancing algorithm has obvious limitations. This paper proposes a dynamic load-balancing model based on the SSAWF (Strong Suspend And Weak Forecast) mechanism. This model uses strong suspend mechanism and cubic exponential smoothing prediction method based on AHP algorithm for dynamic load balancing scheduling. Results of the experiments show that the improved model has more positive influence on performance of the cluster under abnormal system transient performance, high concurrency and high system load interaction, thats to say the load balancing effect is better than the traditional load balancing.

CloudSync: Multi-nodes Directory Synchronization

Directory synchronization aims to maintain the consistency of specified directories. These directories come from different computers and the process of synchronization requires several steps: differences found, differences propagation, conflict discovery, conflict resolution, changes merge. In WAN environment, multi-node directories synchronization has some important features, frequently joining or leaving nodes, concurrent modification, complicate network connection and routing. CloudSync, a multi-nodes directory synchronization system has been proposed. CloudSync compares the current state of directory with last synchronization state to discover the changed files. A master node is used to save and transfer copies of the client changed files. CloudSync proposed a new algorithm which combines the conflict discovery mechanism of SVN with the Vector Clock algorithm to manage the replication files and find conflicts quickly. Eventual consistency between the multi-nodes has been guaranteed.

The optimization research on map marker coverage

The development of computers application changes our daily life. Online maps application and popularization provide us a more convenience life. As individual users, we usually search for single or several map markers on the Internet. For the existing online map, whether from the positioning accuracy, reaction speed, and operating convenience aspects, it is an advanced system. But for the abundant makers, due to the positions of the searched makers are so close, there is a problem of overlapping coverage between markers and markers at a certain zoom level, which requires the maker clustering algorithm to optimize the geographic information display. Through the aided optimize of interaction design in the process of maker clustering, it can help us to enhance the study of geographic data mining and analysis.

MapReduce system productivity measurement model and measuring approach

How to effectively utilize the computing resource remains a longstanding challenge in MapReduce application, and MapReduce system productivity has become a major issue in research field. In the paper, we explored the productivity mathematical models for MapReduce system, defining the productivity as the ratio of the workload and energy consumption per unit time, and proposed the measurement approach for MapReduce system productivity. Based on the measurement model and approach, this paper selected CPU intensive and I/O intensive computing in MapReduce to measure and evaluate MapReduce system productivity. Finally the measurement and evaluation result showed that productivity model and approach in the paper is reasonable and valid for architecture design of MapReduce system in production environment.

An Improved File System Synchronous Algorithm

In order to timely synchronize the nodes in unstable network environment and reduce the space overhead of the operation log, we mix the operation transfer-based and state transfer-based synchronous models. In this paper, for file system synchronization, we design an improved file system synchronous algorithm -- the combination of state transform algorithm and operation merge algorithm. The experimental results show that the improved algorithm can completely merge file system changes into updates operation in millisecond. The key to mix the two ways is to unify updating granularity and semantic level, which will be taken by both models. The mixed model can meet the demand of instant response and the less resource-consuming at the same time.

A Distributed Storage Performance Test System: Design, Implementation, and Experience

with the rapid development of storage technology, many kinds of storage products and systems have been released, but these products and systems always have some problems such as the performance is not credible, the reliability is not high and so on. In order to better test and evaluate the performance of storage products and systems, we design and implement a distributed Storage Performance Test System called SPTS, which can test the IO performance of file system and disk array. The experience of testing disk arrays IOPS and data transfer rate shows that SPTS can provide certain value when customers purchase and use storage products or storage systems.

An Automatic Parameter Optimization Method on Performance Test of Storage Systems

In order to acquire the peak performance of one storage system quickly and accurately, an automatic parameter optimization method is used in performance test. By introducing artificial intelligence into storage test, the algorithms such as particle swarm optimization help to optimize parameters automatically on performance test. Experimental results show that this method can significantly improve efficiency and accuracy of performance test, so saving cost. The method based on intelligent optimization algorithms can acquire the optimal performance value of storage systems more quickly, especially in the case of there are too many parameters in performance test.

MDMeter A New Test Tool for Meta-data Throughput of Parallel File System

As the existing performance evaluation benchmark software for parallel file system has too many test items, it can’t evaluate the meta-data throughput accurately, and there are great difficulties to evaluate the meta-data throughput performance in practical application environment. To solve these problems, a new meta-data throughput evaluation tool, called as MDMeter, has been designed and accomplished in this paper. A new simulation model for file operation is designed according to the real file operation records which were captured by IOTrace tool. MDMeter can launch a variety of meta-data operations on parallel file system, and can provide a comprehensive test result report to users; it can evaluate the meta-data throughput performance accurately for different application environment. At the last, some tests are performed on Lustre parallel file system by MDMeter and traditional tools, the results show that MDMeter not only can give a simple and integrated performance test result to users, but also can evaluate the meta-data throughput for parallel file system more precisely than the existing test tools.

Storage Appliance System Based on Content Addressable Storage

It is difficult for enterprise to build CAS (Content Addressable Storage) appliance system because the most of CAS systems only provide API interface, to design and implement a new kind of CAS storage system is proposed in this paper. CAS interface based on http is suggested, which will turn the operation of file objects into the WEB semantic operation of URI resources, so that the management and operation of file objects can perform through a WEB browser. The file object is described by the metadata model, and CAS client is established with the help of WEB server and database server, so a storage appliance system with powerful capabilities for content navigation and search can be constructed. It is easy to integrate Storage appliance system with enterprise application. The CAS appliance system is extensible; the test results show that aggregate throughput of CAS appliance client will increase greatly with number of CAS server.